Xiangjue Dong, Cong Wang, Maria Teleki, Millennium Bismay, Ruihong Huang, James Caverlee. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Kai Yin, Xiangjue Dong, Chengkai Liu, Allen Lin, Lingfeng Shi, Ali Mostafavi, James Caverlee. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large Language Models (LLMs) excel at capturing latent semantics and contextual relationships across diverse modalities. However, in modeling user behavior from sequential interaction data, performance often suffers when such semantic context is limited or absent. We introduce LaMAR, a LLM-driven semantic enrichment framework designed to enrich such sequences automatically. LaMAR leverages LLMs in a few-shot setting to generate auxiliary contextual signals by inferring latent semantic aspects of a user's intent and item relationships from existing metadata. These generated signals, such as inferred usage scenarios, item intents, or thematic summaries, augment the original sequences with greater contextual depth. We demonstrate the utility of this generated resource by integrating it into benchmark sequential modeling tasks, where it consistently improves performance. Further analysis shows that LLM-generated signals exhibit high semantic novelty and diversity, enhancing the representational capacity of the downstream models. This work represents a new data-centric paradigm where LLMs serve as intelligent context generators, contributing a new method for the semi-automatic creation of training data and language resources.

For decades, the WSDM community has viewed personalization through the lens of collaborative filtering and predictive behavioral models. Today, we are witnessing a fundamental shift where AI systems promise to not just predict user intent, but to actively reason, plan, and create on our behalf. Instead of merely reinforcing our existing habits, these new approaches promise to surface insights we are blind to, guiding us toward discoveries that advance our personal journeys in new and unexpected ways. But are we truly on the verge of super-intelligent personalization?

Building personalized recommender systems and search experiences is a cornerstone of the modern data mining and applied machine learning (ML) community. Modern online platforms have a confluence of data including user-item interaction graphs, user and item-associated semantics (text, visual content, etc.), and metadata. Recent advancements in generative models and semantic encoders via large language models (LLMs), visual and audio encoders have significantly impacted research in relevant domains, enabling new directions in knowledge discovery and ability of models to better incorporate semantic context. These techniques are quickly advancing in the academic sphere, and adoption in industrial environments is growing. These advances force large questions about the future of search, recommendation and personalized experiences in the future. This workshop bridges the research gap between the use of generative models and recommendation for personalized systems. We will focus on topics spanning the interplay between such models and conventional personalized systems. Building upon the momentum of previous successful forums, we seek to engage a diverse audience from academia and industry, fostering a dialogue that incorporates fresh insights and anticipates over 100 attendees, including key stakeholders in the field.

We consider the conditional generation of 3D drug-like molecules with explicit control over molecular properties such as drug-like properties (e.g., Quantitative Estimate of Druglikeness or Synthetic Accessibility score) and effectively binding to specific protein sites. To tackle this problem, we propose an E(3)-equivariant Wasserstein autoencoder and factorize the latent space of our generative model into two disentangled aspects: molecular properties and the remaining structural context of 3D molecules. Our model ensures explicit control over these molecular attributes while maintaining equivariance of coordinate representation and invariance of data likelihood. Furthermore, we introduce a novel alignment-based coordinate loss to adapt equivariant networks for auto-regressive de-novo 3D molecule generation from scratch. Extensive experiments validate our model's effectiveness on property-guided and context-guided molecule generation, both for de-novo 3D molecule design and structure-based drug discovery against protein targets.

, a fact-checking framework that uses extracted knowledge graphs to enhance text representation. Graph Neural Networks further process these graphs as a soft prompt, enabling LLMs to incorporate structured knowledge more effectively. Enhanced with graph-based reasoning, GraphCheck captures multihop reasoning chains that are often overlooked by existing methods, enabling precise and efficient fact-checking in a single inference call. Experimental results on seven benchmarks spanning both general and medical domains demonstrate up to a 7.1% overall improvement over baseline models. Notably, GraphCheck outperforms existing specialized fact-checkers and achieves comparable performance with state-of-the-art LLMs, such as DeepSeek-V3 and OpenAI-o1, with significantly fewer parameters.

Masculine defaults are widely recognized as a significant type of gender bias, but they are often unseen as they are under-researched. Masculine defaults involve three key parts: (i) the cultural context, (ii) the masculine characteristics or behaviors, and (iii) the reward for, or simply acceptance of, those masculine characteristics or behaviors. In this work, we study discourse-based masculine defaults, and propose a twofold framework for (i) the large-scale discovery and analysis of gendered discourse words in spoken content via our Gendered Discourse Correlation Framework (GDCF); and (ii) the measurement of the gender bias associated with these gendered discourse words in LLMs via our Discourse Word-Embedding Association Test (D-WEAT). We focus our study on podcasts, a popular and growing form of social media, analyzing 15,117 podcast episodes. We analyze correlations between gender and discourse words -- discovered via LDA and BERTopic -- to automatically form gendered discourse word lists. We then study the prevalence of these gendered discourse words in domain-specific contexts, and find that gendered discourse-based masculine defaults exist in the domains of business, technology/politics, and video games. Next, we study the representation of these gendered discourse words from a state-of-the-art LLM embedding model from OpenAI, and find that the masculine discourse words have a more stable and robust representation than the feminine discourse words, which may result in better system performance on downstream tasks for men. Hence, men are rewarded for their discourse patterns with better system performance by one of the state-of-the-art language models -- and this embedding disparity is a representational harm and a masculine default.

Generative models have shown great promise in collaborative filtering by capturing the underlying distribution of user interests and preferences. However, existing approaches struggle with inaccurate posterior approximations and misalignment with the discrete nature of recommendation data, limiting their expressiveness and real-world performance. To address these limitations, we propose FlowCF, a novel flow-based recommendation system leveraging flow matching for collaborative filtering. We tailor flow matching to the unique challenges in recommendation through two key innovations: (1) a behavior-guided prior that aligns with user behavior patterns to handle the sparse and heterogeneous user-item interactions, and (2) a discrete flow framework to preserve the binary nature of implicit feedback while maintaining the benefits of flow matching, such as stable training and efficient inference. Extensive experiments demonstrate that FlowCF achieves state-of-the-art recommendation accuracy across various datasets with the fastest inference speed, making it a compelling approach for real-world recommender systems. The code is available at https://github.com/chengkai-liu/FlowCF.

Building personalized recommender systems is a cornerstone of the modern data mining and applied machine learning (ML) community. Modern online platforms have a confluence of data including user-item interaction graphs, user and item-associated semantics (text, visual content, etc.), and metadata. Recent advancements in generative models and semantic encoders via large language models (LLMs), visual and audio encoders have significantly impacted research in relevant domains, enabling new directions in knowledge discovery and ability of models to better incorporate semantic context. This workshop bridges the research gap between the use of generative models and recommendation for personalized systems. We will focus on topics spanning the interplay between such models and conventional personalized systems.

Collaborative Filtering (CF) based recommenders often exhibit model biases, delivering strong recommendation utility to certain users or items at the expense of others. Prior research approaches these biases as isolated and standalone issues, ignoring their interconnected nature and developing separate methods, thereby compromising the specialized debiasing efforts. Thus, we introduce a boosting-based framework designed to alleviate a broad spectrum of biases. This framework employs a series of sub-models, each tailored for different user and item subgroups. Theoretically, our model ensures an exponentially decreasing upper bound on the training loss across all user and item types with increasing boosting iterations. Extensive experiments demonstrate its superior debiasing capabilities against state-of-the-art methods across four model bias types. Appendix, data and code are available at https://github.com/JP-25/CFBoost

Sequential recommender systems aims to predict the users' next interaction through user behavior modeling with various operators like RNNs and attentions. However, existing models generally fail to achieve the three golden principles for sequential recommendation simultaneously, i.e., training efficiency, low-cost inference, and strong performance. To this end, we propose RecBLR, an Efficient Sequential Recommendation Model based on Behavior-Dependent Linear Recurrent Units to accomplish the impossible triangle of the three principles. By incorporating gating mechanisms and behavior-dependent designs into linear recurrent units, our model significantly enhances user behavior modeling and recommendation performance. Furthermore, we unlock the parallelizable training as well as inference efficiency for our model by designing a hardware-aware scanning acceleration algorithm with a customized CUDA kernel. Extensive experiments on real-world datasets with varying lengths of user behavior sequences demonstrate RecBLR's remarkable effectiveness in simultaneously achieving all three golden principles - strong recommendation performance, training efficiency, and low-cost inference, while exhibiting excellent scalability to datasets with long user interaction histories.

Voice-driven software systems are in abundance. However, language models that power these systems are traditionally trained on fluent, written text corpora. Hence there can be a misalignment between the inherent disfluency of transcribed spoken content and the fluency of the written training data. Furthermore, gold-standard disfluency annotations of various complexities for incremental training can be expensive to collect. So, we propose in this paper a Disfluency Augmented Curriculum Learning (DACL) approach to tackle the complex structure of disfluent sentences and generate fluent texts from them, by using Curriculum Learning (CL) coupled with our synthetically augmented disfluent texts of various levels. DACL harnesses the tiered structure of our generated synthetic disfluent data using CL, by training the model on basic samples (i.e. more fluent) first before training it on more complex samples (i.e. more disfluent). In contrast to the random data exposure paradigm, DACL focuses on a simple-to-complex learning process. We comprehensively evaluate DACL on Switchboard Penn Treebank-3 and compare it to the state-of-the-art disfluency removal models. Our model surpasses existing techniques in word-based precision (by up to 1%) and has shown favorable recall and F1 scores.

Spoken content is abundant – including podcasts, meeting transcripts, and TikTok-like short videos. And yet, many important tasks like summarization are often designed for written content rather than the looser, noiser, and more disfluent style of spoken content. Hence, we aim in this paper to quantify the impact of disfluency on spoken content summarization. Do disfluencies negatively impact the quality of summaries generated by existing approaches? And if so, to what degree? Coupled with these goals, we also investigate two methods towards improving summarization in the presence of such disfluencies. We find that summarization quality does degrade with an increase in these disfluencies and that a combination of multiple disfluency types leads to even greater degradation. Further, our experimental results show that naively removing disfluencies and augmenting with special tags can worsen the summarization when used for testing, but that removing disfluencies for fine-tuning yields the best results. We make the code available at https://github.com/mariateleki/Quantifying-Impact-Disfluency.

Vision-language models (VLMs) excel in zero-shot recognition but their performance varies greatly across different visual concepts. For example, although CLIP achieves impressive accuracy on ImageNet (60-80%), its performance drops below 10% for more than ten concepts like night snake, presumably due to their limited presence in the pretraining data. However, measuring the frequency of concepts in VLMs' large-scale datasets is challenging. We address this by using large language models (LLMs) to count the number of pretraining texts that con-tain synonyms of these concepts. Our analysis confirms that popular datasets, such as LAION, exhibit a long-tailed concept distribution, yielding biased performance in VLMs. We also find that downstream applications of VLMs, including visual chatbots (e.g., GPT-4V) and text-to-image models (e.g., Stable Diffusion), often fail to recognize or generate images of rare concepts identified by our method. To mit-igate the imbalanced performance of zero-shot VLMs, we propose REtrieval-Augmented Learning (REAL). First, in-stead of prompting VLMs using the original class names, REAL uses their most frequent synonyms found in pretraining texts. This simple change already outperforms costly human-engineered and LLM-enriched prompts over nine benchmark datasets. Second, REAL trains a linear classifier on a small yet balanced set of pretraining data re-trieved using concept synonyms. REAL surpasses the previous zero-shot SOTA, using 400× less storage and 10,000× less training time!

Guanchu Wang, Yu-Neng Chuang, Ruixiang Tang, Shaochen Zhong, Jiayi Yuan, Hongye Jin, Zirui Liu, Vipin Chaudhary, Shuai Xu, James Caverlee, Xia Hu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Language models can be manipulated by adversarial attacks, which introduce subtle perturbations to input data.While recent attack methods can achieve a relatively high attack success rate (ASR), we've observed that the generated adversarial examples have a different data distribution compared with the original examples.Specifically, these adversarial examples exhibit reduced confidence levels and greater divergence from the training data distribution.Consequently, they are easy to detect using straightforward detection methods, diminishing the efficacy of such attacks.To address this issue, we propose a Distribution-Aware Adversarial Attack (DA 3 ) method.DA 3 considers the distribution shifts of adversarial examples to improve attacks' effectiveness under detection methods.We further design a novel evaluation metric, the Non-detectable Attack Success Rate (NASR), which integrates both ASR and detectability for the attack task.We conduct experiments on four widely used datasets to validate the attack effectiveness and transferability of adversarial examples generated by DA 3 against both the white-box BERT-BASE and ROBERTA-BASE models and the black-box LLAMA2-7B model 1 .

In recent years, massive transformer-based architectures have driven breakthrough performance in practical applications like autoregressive text-generation (LLMs) and click-prediction (recommenders). A common recipe for success is to train large models on massive web-scale datasets [3, 15], e.g., modern recommenders are trained on billions of user-item click events, and LLMs are trained on trillions of tokens extracted from the public internet. We are close to hitting the computational and economical limits of scaling up the size of these models, and we expect the next frontier of gains to come from improving the: (i) data quality of the training dataset, and (ii) data efficiency of the extremely expensive training procedure. Inspired by this shift, we present a set of “data-centric” techniques for recommendation and language models that summarizes a dataset into a terse data summary, which is both (i) high-quality, i.e., trains better quality models, and (ii) improves the data-efficiency of the overall training procedure. We propose techniques from two disparate data frameworks: (i) data selection (a.k.a., coreset construction) methods that sample portions of the dataset using grounded heuristics, and (ii) data distillation techniques that generate synthetic examples which are optimized to retain the signals needed for training high-quality models. Overall, this work sheds light on the challenges and opportunities offered by data optimization in web-scale systems, a particularly relevant focus as the recommendation community grapples with the grand challenge of leveraging LLMs.

Graph machine learning (Graph ML) models typically require abundant labeled instances to provide sufficient supervision signals, which is commonly infeasible in real-world scenarios since labeled data for newly emerged concepts (e.g., new categorizations of nodes) on graphs is rather limited. To efficiently learn with a small amount of data on graphs, meta-learning has been investigated in Graph ML. By transferring the knowledge learned from previous experiences to new tasks, graph meta-learning approaches have demonstrated promising performance on few-shot graph learning problems. However, most existing efforts predominately assume that all the data from the seen classes is gold labeled, yet those methods may lose their efficacy when the seen data is weakly labeled with severe label noise. As such, we aim to investigate a novel problem of weakly supervised graph meta-learning for improving the model robustness in terms of knowledge transfer. To achieve this goal, we propose Meta-GIN (Meta Graph Interpolation Network), a new graph meta-learning framework. Based on a new robustness-enhanced episodic training paradigm, Meta-GIN is meta-learned to interpolate node representations from weakly labeled data and extracts highly transferable meta-knowledge, which enables the model to quickly adapt to unseen tasks with few labeled instances. Extensive experiments demonstrate the superiority of Meta-GIN over existing graph meta-learning studies on the task of weakly supervised few-shot node classification.

Question Generation (QG) is a fundamental NLP task for many downstream applications.Recent studies on open-book QG, where supportive answer-context pairs are provided to models, have achieved promising progress.However, generating natural questions under a more practical closed-book setting that lacks these supporting documents still remains a challenge.In this work, we propose a new QG model for this closed-book setting that is designed to better understand the semantics of long-form abstractive answers and store more information in its parameters through contrastive learning and an answer reconstruction module.Through experiments, we validate the proposed QG model on both public datasets and a new WikiCQA dataset.Empirical results show that the proposed QG model outperforms baselines in both automatic evaluation and human evaluation.In addition, we show how to leverage the proposed model to improve existing question-answering systems.These results further indicate the effectiveness of our QG model for enhancing closed-book questionanswering tasks.

Subjective bias is ubiquitous on news sites, social media, and knowledge resources like Wikipedia.Many existing methods for subjective bias correction have typically focused on making one-word edits and have been trained over a single (often, noisy) domain.In contrast, we propose a novel reinforced sequence training approach for robust subjective bias correction.Three of the unique characteristics of the approach are: (i) it balances bias neutralization with fluency and semantics preservation through reinforcement learning, to broaden the scope to bias beyond a single word; (ii) it is cross-trained over multiple sources of bias to be more robust to new styles of biased writing that are not seen in the training data for a single domain; and (iii) it is used to fine-tune a large pre-trained transformer model to yield state-ofthe-art performance in bias text correction task.Extensive experiments show that the proposed approach results in significant improvements in subjective bias correction versus alternatives.

Objectivity is a goal for Wikipedia and many news sites, as well as a guiding principle of many large language models.Indeed, several methods have recently been developed for automatic subjective bias neutralization.These methods, however, typically rely on parallel text for training (i.e. a biased sentence coupled with a non-biased sentence), demonstrate poor transfer to new domains, and can lose important bias-independent context.Toward expanding the reach of bias neutralization, we propose in this paper a new approach called FairBalance.Three of its unique features are: i) a cycle consistent adversarial network enables bias neutralization without the need for parallel text; ii) the model design preserves bias-independent content; and iii) through auxiliary guidance, the model highlights sequences of bias-inducing words, yielding strong results in terms of bias neutralization quality.In our evaluations involving seven models comprising of adversarial and non-adversarial models, the FairBalance method showed a notable improvement in bias neutralization based on subjective human judgment when compared to other techniques.

Industry recommender systems usually suffer from highly-skewed long-tail item distributions where a small fraction of the items receives most of the user feedback. This skew hurts recommender quality especially for the item slices without much user feedback. While there have been many research advances made in academia, deploying these methods in production is very difficult and very few improvements have been made in industry. One challenge is that these methods often hurt overall performance; additionally, they could be complex and expensive to train and serve.

Sequential models are designed to learn sequential patterns in data based on the chronological order of user interactions. However, they often ignore the timestamps of these interactions. Incorporating time is crucial because many sequential patterns are time-dependent, and the model cannot make time-aware recommendations without considering time. This article demonstrates that providing a rich representation of time can significantly improve the performance of sequential models. The existing literature treats time as a one-dimensional time-series obtained by quantizing time. In this study, we propose treating time as a multi-dimensional time-series and explore representation learning methods, including a kernel based method and an embedding-based algorithm. Experiments on multiple datasets show that the inclusion of time significantly enhances the model’s performance, and multi-dimensional methods outperform the one-dimensional method by a substantial margin.

Conversational recommenders are emerging as a powerful tool to personalize a user’s recommendation experience. Through a back-and-forth dialogue, users can quickly hone in on just the right items. Many approaches to conversational recommendation, however, only partially explore the user preference space and make limiting assumptions about how user feedback can be best incorporated, resulting in long dialogues and poor recommendation performance. In this paper, we propose a novel conversational recommendation framework with two unique features: (i) a greedy NDCG attribute selector, to enhance user personalization in the interactive preference elicitation process by prioritizing attributes that most effectively represent the actual preference space of the user; and (ii) a user representation refiner, to effectively fuse together the user preferences collected from the interactive elicitation process to obtain a more personalized understanding of the user. Through extensive experiments on four frequently used datasets, we find the proposed framework not only outperforms all the state-of-the-art conversational recommenders (in terms of both recommendation performance and conversation efficiency), but also provides a more personalized experience for the user under the proposed multi-groundtruth multi-round conversational recommendation setting.

Inspired by the extensive success of deep learning, graph neural networks (GNNs) have been proposed to learn expressive node representations and demonstrated promising performance in various graph learning tasks. However, existing endeavors predominately focus on the conventional semi-supervised setting where relatively abundant gold-labeled nodes are provided. While it is often impractical due to the fact that data labeling is unbearably laborious and requires intensive domain knowledge, especially when considering the heterogeneity of graph-structured data. Under the few-shot semi-supervised setting, the performance of most of the existing GNNs is inevitably undermined by the overfitting and oversmoothing issues, largely owing to the shortage of labeled data. In this paper, we propose a decoupled network architecture equipped with a novel meta-learning algorithm to solve this problem. In essence, our framework Meta-PN infers high-quality pseudo labels on unlabeled nodes via a meta-learned label propagation strategy, which effectively augments the scarce labeled data while enabling large receptive fields during training. Extensive experiments demonstrate that our approach offers easy and substantial performance gains compared to existing techniques on various benchmark datasets. The implementation and extended manuscript of this work are publicly available at https://github.com/kaize0409/Meta-PN.

Conversational recommender systems (CRS) have shown great success in accurately capturing a user's current and detailed preference through the multi-round interaction cycle while effectively guiding users to a more personalized recommendation. Perhaps surprisingly, conversational recommender systems can be plagued by popularity bias, much like traditional recommender systems. In this paper, we systematically study the problem of popularity bias in CRSs. We demonstrate the existence of popularity bias in existing state-of-the-art CRSs from an exposure rate, a success rate, and a conversational utility perspective, and propose a suite of popularity bias metrics designed specifically for the CRS setting. We then introduce a debiasing framework with three unique features: (i) Popularity-Aware Focused Learning, to reduce the popularity-distorting impact on preference prediction; (ii) Cold-Start Item Embedding Reconstruction via Attribute Mapping, to improve the modeling of cold-start items; and (iii) Dual-Policy Learning, to better guide the CRS when dealing with either popular or unpopular items. Through extensive experiments on two frequently used CRS datasets, we find the proposed model-agnostic debiasing framework not only mitigates the popularity bias in state-of-the-art CRSs but also improves the overall recommendation performance.

In collaborative filtering, the quality of recommendations critically relies on how easily a model can find similar users for a target user. Hence, a niche user who prefers items out of the mainstream may receive poor recommendations, while a mainstream user sharing interests with many others will likely receive recommendations of higher quality. In this work, we study this mainstream bias centering around three key thrusts. First, to distinguish mainstream and niche users, we explore four approaches based on outlier detection techniques to identify a mainstream score indicating the mainstream level for each user. Second, we empirically show that severe mainstream bias is produced by conventional recommendation models. Last, we explore both global and local methods to mitigate the bias. Concretely, we propose two global models: Distribution Calibration (DC) and Weighted Loss (WL) methods; and one local method: Local Fine Tuning (LFT) method. Extensive experiments show the effectiveness of the proposed methods to improve utility for niche users and also show that the proposed LFT can improve the utility for mainstream users at the same time.

In many personalized recommendation scenarios, the generalization ability of a target task can be improved via learning with additional auxiliary tasks alongside this target task on a multi-task network. However, this method often suffers from a serious optimization imbalance problem. On the one hand, one or more auxiliary tasks might have a larger influence than the target task and even dominate the network weights, resulting in worse recommendation accuracy for the target task. On the other hand, the influence of one or more auxiliary tasks might be too weak to assist the target task. More challenging is that this imbalance dynamically changes throughout the training process and varies across the parts of the same network. We propose a new method: MetaBalance to balance auxiliary losses via directly manipulating their gradients w.r.t the shared parameters in the multi-task network. Specifically, in each training iteration and adaptively for each part of the network, the gradient of an auxiliary loss is carefully reduced or enlarged to have a closer magnitude to the gradient of the target loss, preventing auxiliary tasks from being so strong that dominate the target task or too weak to help the target task. Moreover, the proximity between the gradient magnitudes can be flexibly adjusted to adapt MetaBalance to different scenarios. The experiments show that our proposed method achieves a significant improvement of 8.34% in terms of [email protected] upon the strongest baseline on two real-world datasets. The code of our approach can be found at here.1

In a modern e-commerce recommender system, it is important to understand the relationships among products. Recognizing product relationships-such as complements or substitutes-accurately is an essential task for generating better recommendation results, as well as improving explainability in recommendation. Products and their associated relationships naturally form a product graph, yet existing efforts do not fully exploit the product graph's topological structure. They usually only consider the information from directly connected products. In fact, the connectivity of products a few hops away also contains rich semantics and could be utilized for improved relationship prediction. In this work, we formulate the problem as a multilabel link prediction task and propose a novel graph neural network-based framework, item relationship graph neural network (IRGNN), for discovering multiple complex relationships simultaneously. We incorporate multihop relationships of products by recursively updating node embeddings using the messages from their neighbors. An edge relational network is designed to effectively capture relational information between products. Extensive experiments are conducted on real-world product data, validating the effectiveness of IRGNN, especially on large and sparse product graphs.

Social Resonance is a common socio-behavioral phenomenon in which users are more influenced by opinions that have similar vibes. That is, opinions from two different groups of users can mutually reinforce (or resonate with) each other to have an even stronger impact on the user. In this paper, we explore the powerful social resonance effect between social connections and other users in an eCommerce platform to improve recommendation. Specifically, we first formulate an item-aware user influence network that connects users who rate the same item. With the social network and item-aware user influence network, a novel graph-based mutual learning framework is proposed, which captures the resonance influence from both user local correlations and global connections. We then fuse these influence paths to predict the resonance-enhanced user preference towards items. Experiments on public benchmarks show the proposed approach outperforms state-of-the-art social recommendation methods.

Popularity bias is a long-standing challenge in recommender systems: popular items are overly recommended at the expense of less popular items that users may be interested in being under-recommended. Such a bias exerts detrimental impact on both users and item providers, and many efforts have been dedicated to studying and solving such a bias. However, most existing works situate the popularity bias in a static setting, where the bias is analyzed only for a single round of recommendation with logged data. These works fail to take account of the dynamic nature of real-world recommendation process, leaving several important research questions unanswered: how does the popularity bias evolve in a dynamic scenario? what are the impacts of unique factors in a dynamic recommendation process on the bias? and how to debias in this long-term dynamic process? In this work, we investigate the popularity bias in dynamic recommendation and aim to tackle these research gaps. Concretely, we conduct an empirical study by simulation experiments to analyze popularity bias in the dynamic scenario and propose a dynamic debiasing strategy and a novel False Positive Correction method utilizing false positive signals to debias, which show effective performance in extensive experiments.

Session-based recommender systems aim to improve recommendations in short-term sessions that can be found across many platforms.A critical challenge is to accurately model user intent with only limited evidence in these short sessions.For example, is a flower bouquet being viewed meant as part of a wedding purchase or for home decoration?Such different perspectives greatly impact what should be recommended next.Hence, this paper proposes a novel sessionbased recommendation system empowered by hypergraph attention networks.Three unique properties of the proposed approach are: (i) it constructs a hypergraph for each session to model the item correlations defined by various contextual windows in the session simultaneously, to uncover item meanings; (ii) it is equipped with hypergraph attention layers to generate item embeddings by flexibly aggregating the contextual information from correlated items in the session; and (iii) it aggregates the dynamic item representations for each session to infer the general purpose and current need, which is decoded to infer the next interesting item in the session.Through experiments on three benchmark datasets, we find the proposed model is effective in generating informative dynamic item embeddings and providing more accurate recommendations compared to the state-of-the-art.

A fundamental challenge for sequential recommenders is to capture the sequential patterns of users toward modeling how users transit among items. In many practical scenarios, however, there are a great number of cold-start users with only minimal logged interactions. As a result, existing sequential recommendation models will lose their predictive power due to the difficulties in learning sequential patterns over users with only limited interactions. In this work, we aim to improve sequential recommendation for cold-start users with a novel framework named MetaTL, which learns to model the transition patterns of users through meta-learning. Specifically, the proposed MetaTL: (i) formulates sequential recommendation for cold-start users as a few-shot learning problem; (ii) extracts the dynamic transition patterns among users with a translation-based architecture; and (iii) adopts meta transitional learning to enable fast learning for cold-start users with only limited interactions, leading to accurate inference of sequential interactions.

This paper investigates recommendation fairness among new items. While previous efforts have studied fairness in recommender systems and shown success in improving fairness, they mainly focus on scenarios where unfairness arises due to biased prior user-feedback history (like clicks or views). Yet, it is unknown whether new items without any feedback history can be recommended fairly, and if unfairness does exist, how can we provide fair recommendations among these new items in such a cold-start scenario. In detail, we first formalize fairness among new items with the well-known concepts of equal opportunity and Rawlsian Max-Min fairness. We empirically show the prevalence of unfairness in cold start recommender systems. Then we propose a novel learnable post-processing framework as a model blueprint for enhancing fairness, with which we propose two concrete models: a joint-learning generative model, and a score scaling model. Extensive experiments over four public datasets show the effectiveness of the proposed models for enhancing fairness while also preserving recommendation utility.

This paper connects equal opportunity to popularity bias in implicit recommenders to introduce the problem of popularity-opportunity bias. That is, conditioned on user preferences that a user likes both items, the more popular item is more likely to be recommended (or ranked higher) to the user than the less popular one. This type of bias is harmful, exerting negative effects on the engagement of both users and item providers. Thus, we conduct a three-part study: (i) By a comprehensive empirical study, we identify the existence of the popularity-opportunity bias in fundamental matrix factorization models on four datasets; (ii) coupled with this empirical study, our theoretical study shows that matrix factorization models inherently produce the bias; and (iii) we demonstrate the potential of alleviating this bias by both in-processing and post-processing algorithms. Extensive experiments on four datasets show the effective debiasing performance of these proposed methods compared with baselines designed for conventional popularity bias.

To mitigate the rabbit hole effect in recommendations, conventional distribution-aware recommendation systems aim to ensure that a user’s prior interest areas are reflected in the recommendations that the system makes. For example, a user who historically prefers comedies to dramas by 2:1 should see a similar ratio in recommended movies. Such approaches have proven to be an important building block for recommendation tasks. However, existing distribution-aware approaches enforce that the target taste distribution should exactly match a user’s prior interests (typically revealed through training data), based on the assumption that users’ taste distribution is fundamentally static. This assumption can lead to large estimation errors. We empirically identify this taste distortion problem through a data-driven study over multiple datasets. We show how taste preferences dynamically shift and how the design of a calibration mechanism should be designed with these shifts in mind. We further demonstrate how to incorporate these shifts into a taste enhanced calibrated recommender system, which results in simultaneously mitigated both the rabbit hole effect and taste distortion problem.

We present a new benchmark dataset called PARADE for paraphrase identification that requires specialized domain knowledge.PA-RADE contains paraphrases that overlap very little at the lexical and syntactic level but are semantically equivalent based on computer science domain knowledge, as well as nonparaphrases that overlap greatly at the lexical and syntactic level but are not semantically equivalent based on this domain knowledge.Experiments show that both state-of-the-art neural models and non-expert human annotators have poor performance on PARADE.For example, BERT after fine-tuning achieves an F1 score of 0.709, which is much lower than its performance on other paraphrase identification datasets.PARADE can serve as a resource for researchers interested in testing models that incorporate domain knowledge.We make our data and code freely available.1

Knowledge of a disease includes information of various aspects of the disease, such as signs and symptoms, diagnosis and treatment.This disease knowledge is critical for many healthrelated and biomedical tasks, including consumer health question answering, medical language inference and disease name recognition.While pre-trained language models like BERT have shown success in capturing syntactic, semantic, and world knowledge from text, we find they can be further complemented by specific information like knowledge of symptoms, diagnoses, treatments, and other disease aspects.Hence, we integrate BERT with disease knowledge for improving these important tasks.Specifically, we propose a new disease knowledge infusion training procedure and evaluate it on a suite of BERT models including BERT, BioBERT, SciBERT, Clinical-BERT, BlueBERT, and ALBERT.Experiments over the three tasks show that these models can be enhanced in nearly all cases, demonstrating the viability of disease knowledge infusion.For example, accuracy of BioBERT on consumer health question answering is improved from 68.29% to 72.09%, while new SOTA results are observed in two datasets.We make our data and code freely available.

Modern recommenders usually consider both collaborative features from user behavior data (e.g., clicks) and content information about the users and items (e.g., user ages or item images) for improved recommendations. While encouraging, the uncovered user preference representations derived from these collaborative and content-based perspectives can be entangled by intermixing the influence from each other, leading to sub-optimal performance and unstable recommendations. Hence, we propose to disentangle representations learned from user behavior data and content information. Specifically, we propose a novel two-level disentanglement generative recommendation model (DICER) that supports both content-collaborative disentanglement and feature disentanglement: for the content-collaborative disentanglement, DICER decomposes the features by their marginal distributions based on content and user-item interactions, to ensure the learned features from each type are statistically independent. For feature disentanglement, by decomposing the Kullback-Leibler divergence, we theoretically show that extracted features within each type are disentangled at a granular level. Furthermore, DICER utilizes a co-decoder that simultaneously decodes the content and user-item interactions to ensure the high-quality of learned features. Through extensive experiments on three real-world datasets, results show that DICER significantly outperforms other state-of-the-art methods by 13.5% in NDCG and 14.4% in hit ratio on average.

This paper focuses on how to generate unbiased recommendations based on biased implicit user-item interactions. We propose a combinational joint learning framework to simultaneously learn unbiased user-item relevance and unbiased propensity. More specifically, we first present a new unbiased objective function for estimating propensity. We then show how a naïve joint learning approach faces an estimation-training overlap problem. Hence, we propose to jointly train multiple sub-models from different parts of the training dataset to avoid this problem. Finally, we show how to incorporate residual components trained by the complete training data to complement the relevance and propensity sub-models. Extensive experiments on two public datasets demonstrate the effectiveness of the proposed model with an improvement of 4% on average over the best alternatives.

Online reviews play a critical role in persuading or dissuading users when making purchase decisions. And yet very few users take the time to write helpful reviews. Encouragingly, recent advances in deep neural networks offer good potential to produce review-like natural language content. However, there is a lack of large, high-quality labeled data at both the aspect and sentiment level for training. Hence, toward enabling a writing assistant framework to help users post online reviews, this paper proposes a scalable labeling method for bootstrapping aspect and sentiment labels.Concretely, the proposed approach ? Aspect Dependent Online RE-views (ADORE) - leverages the underlying distribution of reviews and a small seed set of labeled data through carefully designed review segmentation and label assignment. We then show how these labels can inform a generative model to produce aspect and sentiment-aware reviews. We study the effectiveness of ADORE under various scenarios such as how end-users perceive the quality of the labels and aspect-aware generated reviews. Our experiments indicate that the proposed effective labeling process along with a regularized joint generative model lead to high quality reviews with 90% accuracy.

There is an increasing attention on next-item recommendation systems to infer the dynamic user preferences with sequential user interactions. While the semantics of an item can change over time and across users, the item correlations defined by user interactions in the short term can be distilled to capture such change, and help in uncovering the dynamic user preferences. Thus, we are motivated to develop a novel next-item recommendation framework empowered by sequential hypergraphs. Specifically, the framework: (i) adopts hypergraph to represent the short-term item correlations and applies multiple convolutional layers to capture multi-order connections in the hypergraph; (ii) models the connections between different time periods with a residual gating layer; and (iii) is equipped with a fusion layer to incorporate both the dynamic item embedding and short-term user intent to the representation of each interaction before feeding it into the self-attention layer for dynamic user modeling. Through experiments on datasets from the ecommerce sites Amazon and Etsy and the information sharing platform Goodreads, the proposed model can significantly outperform the state-of-the-art in predicting the next interesting item for each user.

The cold start problem is a long-standing challenge in recommender systems. That is, how to recommend for new users and new items without any historical interaction record? Recent ML-based approaches have made promising strides versus traditional methods. These ML approaches typically combine both user-item interaction data of existing warm start users and items (as in CF-based methods) with auxiliary information of users and items such as user profiles and item content information (as in content-based methods). However, such approaches face key drawbacks including the error superimposition issue that the auxiliary-to-CF transformation error increases the final recommendation error; the ineffective learning issue that long distance from transformation functions to model output layer leads to ineffective model learning; and the unified transformation issue that applying the same transformation function for different users and items results in poor transformation.

Recommendation algorithms typically build models based on user-item interactions (e.g., clicks, likes, or ratings) to provide a personalized ranked list of items. These interactions are often distributed unevenly over different groups of items due to varying user preferences. However, we show that recommendation algorithms can inherit or even amplify this imbalanced distribution, leading to item under-recommendation bias. Concretely, we formalize the concepts of ranking-based statistical parity and equal opportunity as two measures of item under-recommendation bias. Then, we empirically show that one of the most widely adopted algorithms -- Bayesian Personalized Ranking -- produces biased recommendations, which motivates our effort to propose the novel debiased personalized ranking model. The debiased model is able to improve the two proposed bias metrics while preserving recommendation performance. Experiments on three public datasets show strong bias reduction of the proposed model versus state-of-the-art alternatives.

User-generated item lists are popular on many platforms. Examples include video-based playlists on YouTube, image-based lists (or "boards") on Pinterest, book-based lists on Goodreads, and answer-based lists on question-answer forums like Zhihu. As users create these lists, a common challenge is in identifying what items to curate next. Some lists are organized around particular genres or topics, while others are seemingly incoherent, reflecting individual preferences for what items belong together. Furthermore, this heterogeneity in item consistency may vary from platform to platform, and from sub-community to sub-community. Hence, this paper proposes a generalizable approach for user-generated item list continuation. Complementary to methods that exploit specific content patterns (e.g., as in song-based playlists that rely on audio features), the proposed approach models the consistency of item lists based on human curation patterns, and so can be deployed across a wide range of varying item types (e.g., videos, images, books). A key contribution is in intelligently combining two preference models via a novel consistency-aware gating network -- a general user preference model that captures a user's overall interests, and a current preference priority model that captures a user's current (as of the most recent item) interests. In this way, the proposed consistency-aware recommender can dynamically adapt as user preferences evolve. Evaluation over four datasets (of songs, books, and answers) confirms these observations and demonstrates the effectiveness of the proposed model versus state-of-the-art alternatives. Further, all code and data are available at https://github.com/heyunh2015/ListContinuation_WSDM2020.

Recommendation systems typically rely on the interactions between a crowd of ordinary users and items, ignoring the fact that many real-world communities are notably influenced by a small group of key opinion leaders, whose feedback on items wields outsize influence. With important positions in the community (e.g. have a large number of followers), their elite opinions are able to diffuse to the community and further impact what items we buy, what media we consume, and how we interact with online platforms. Hence, this paper investigates how to develop a novel recommendation system by explicitly capturing the influence from key opinion leaders to the whole community. Centering around opinion elicitation and diffusion, we propose an end-to-end Graph-based neural model - GoRec. Specifically, to preserve the multi-relations between key opinion leaders and items, GoRec elicits the opinions from key opinion leaders with a translation-based embedding method. Moreover, GoRec adopts the idea of Graph Neural Networks to model the elite opinion diffusion process for improved recommendation. Through experiments on Goodreads and Epinions, the proposed model outperforms state-of-the-art approaches by 10.75% and 9.28% on average in Top-K item recommendation.

Currently, most sequence-based recommendation models aim to predict a user's next actions (e.g. next purchase) based on their past actions. These models either capture users' intrinsic preference (e.g. a comedy lover, or a fan of fantasy) from their long-term behavior patterns or infer their current needs by emphasizing recent actions. However, in e-commerce, intrinsic user behavior may be shifted by occasions such as birthdays, anniversaries, or gifting celebrations (Valentine's Day or Mother's Day), leading to purchases that deviate from long-term preferences and are not related to recent actions. In this work, we propose a novel next-item recommendation system which models a user's default, intrinsic preference, as well as two different kinds of occasion-based signals that may cause users to deviate from their normal behavior. More specifically, this model is novel in that it: (1) captures a personal occasion signal using an attention layer that models reoccurring occasions specific to that user (e.g. a birthday); (2) captures a global occasion signal using an attention layer that models seasonal or reoccurring occasions for many users (e.g. Christmas); (3) balances the user's intrinsic preferences with the personal and global occasion signals for different users at different timestamps with a gating layer. We explore two real-world e-commerce datasets (Amazon and Etsy) and show that the proposed model outperforms state-of-the-art models by 7.62% and 6.06% in predicting users' next purchase.

We propose a personalized user recommendation framework for content curation platforms that models preferences for both users and the items they engage with simultaneously. In this way, user preferences for specific item types (e.g., fantasy novels) can be balanced with user specialties (e.g., reviewing novels with strong female protagonists). In particular, the proposed model has three unique characteristics: (i) it simultaneously learns both user-item and user-user preferences through a multi-aspect autoencoder model; (ii) it fuses the latent representations of user preferences on users and items to construct shared factors through an adversarial framework; and (iii) it incorporates an attention layer to produce weighted aggregations of different latent representations, leading to improved personalized recommendation of users and items. Through experiments against state-of-the-art models, we find the proposed framework leads to a 18.43% (Goodreads) and 6.14% (Spotify) improvement in top-k user recommendation.

The importance of the distribution of ratings on recommender systems (RS) is well-recognized. And yet, recommendation approaches based on latent factor models and recently introduced neural variants (e.g., NCF) optimize for the head of these distributions, potentially leading to large estimation errors for tail ratings. These errors in tail ratings that are far from the mean predicted rating fall out of a uni-modal assumption underlying these popular models, as we show in this paper. We propose to improve the estimation of tail ratings by extending traditional single latent representations (e.g., an item is represented by a single latent vector) with new multi-latent representations for better modeling these tail ratings. We show how to incorporate these multi-latent representations in an end-to-end neural prediction model that is designed to better reflect the underlying ratings distributions of items. Through experiments over six datasets, we find the proposed model leads to a significant improvement in RMSE versus a suite of benchmark methods. We also find that the predictions for the most polarized items are improved by more than 15%.

We propose an adaptive hierarchical translation-based sequential recommendation called HierTrans that first extends traditional item-level relations to the category-level, to help capture dynamic sequence patterns that can generalize across users and time. Then unlike item-level based methods, we build a novel hierarchical temporal graph that contains item multi-relations at the category-level and user dynamic sequences at the item-level. Based on the graph, HierTrans adaptively aggregates the high-order multi-relations among items and dynamic user preferences to capture the dynamic joint influence for next-item recommendation. Specifically, the user translation vector in HierTrans can adaptively change based on both a user’s previous interacted items and the item relations inside the user’s sequences, as well as the user’s personal dynamic preference. Experiments on public datasets demonstrate the proposed model HierTrans consistently outperforms state-of-the-art sequential recommendation methods.

Predicting the potential target customers for a product is essential. However, traditional recommender systems typically aim to optimize an engagement metric without considering the overall distribution of target customers, thereby leading to serious distortion problems. In this paper, we conduct a data-driven study to reveal several distortions that arise from conventional recommenders. Toward overcoming these issues, we propose a target customer re-ranking algorithm to adjust the population distribution and composition in the Top-k target customers of an item while maintaining recommendation quality. By applying this proposed algorithm onto a real-world dataset, we find the proposed method can effectively make the class distribution of items’ target customers close to the desired distribution, thereby mitigating distortion.

User-generated item lists are a popular feature of many different platforms. Examples include lists of books on Goodreads, playlists on Spotify and YouTube, collections of images on Pinterest, and lists of answers on question-answer sites like Zhihu. Recommending item lists is critical for increasing user engagement and connecting users to new items, but many approaches are designed for the item-based recommendation, without careful consideration of the complex relationships between items and lists. Hence, in this paper, we propose a novel user-generated list recommendation model called AttList. Two unique features of AttList are careful modeling of (i) hierarchical user preference, which aggregates items to characterize the list that they belong to, and then aggregates these lists to estimate the user preference, naturally fitting into the hierarchical structure of item lists; and (ii) item and list consistency, through a novel self-attentive aggregation layer designed for capturing the consistency of neighboring items and lists to better model user preference. Through experiments over three real-world datasets reflecting different kinds of user-generated item lists, we find that AttList results in significant improvements in NDCG, [email protected], and [email protected] versus a suite of state-of-the-art baselines. Furthermore, all code and data are available at https://github.com/heyunh2015/AttList.

User reviews have become a cornerstone of how we make decisions. However, this user-based feedback is susceptible to manipulation as recent research has shown the feasibility of automatically generating fake reviews. Previous investigations, however, have focused on generative fake review approaches that are (i) domain dependent and not extendable to other domains without replicating the whole process from scratch; and (ii) character-level based known to generate reviews of poor quality that are easily detectable by anti-spam detectors and by end users. In this work, we propose and evaluate a new class of attacks on online review platforms based on neural language models at word-level granularity in an inductive transfer-learning framework wherein a universal model is refined to handle domain shift, leading to potentially wide-ranging attacks on review systems. Through extensive evaluation, we show that such model-generated reviews can bypass powerful anti-spam detectors and fool end users. Paired with this troubling attack vector, we propose a new defense mechanism that exploits the distributed representation of these reviews to detect model-generated reviews. We conclude that despite the success of neural models in generating realistic reviews, our proposed RNN-based discriminator can combat this type of attack effectively (90% accuracy).

Fashion-focused key opinion bloggers on Instagram, Facebook, and other social media platforms are fast becoming critical influencers. They can inspire consumer clothing purchases by linking high fashion visual evolution with daily street style. In this paper, we build thefirst visual influence-aware fashion recommender (FIRN) with leveraging fashion bloggers and their dynamic visual posts. Specifically, we extract thedynamic fashion features highlighted by these bloggers via a BiLSTM that integrates a large corpus of visual posts and community influence. We then learn theimplicit visual influence funnel from bloggers to individual users via a personalized attention layer. Finally, we incorporate user personal style and her preferred fashion features across time in a recurrent recommendation network for dynamic fashion-updated clothing recommendation. Experiments show that FIRN outperforms state-of-the-art fashion recommenders, especially for users who are most impacted by fashion influencers, and utilizing fashion bloggers can bring greater improvements in recommendation compared with using other potential sources of visual information. We also release a largetime-aware high-quality visual dataset of fashion influencers that can be exploited for future research.

We propose a new time-dependent predictive model of user-item ratings centered around local coherence -- that is, while both users and items are constantly in flux, within a short-term sequence, the neighborhood of a particular user or item is likely to be coherent. Three unique characteristics of the framework are: (i) it incorporates both implicit and explicit feedbacks by extracting the local coherence hidden in the feedback sequences; (ii) it uses parallel recurrent neural networks to capture the evolution of users and items, resulting in a dual factor recommendation model; and (iii) it combines both coherence-enhanced consistent latent factors and dynamic latent factors to balance short-term changes with long-term trends for improved recommendation. Through experiments on Goodreads and Amazon, we find that the proposed model can outperform state-of-the-art models in predicting users' preferences.

In this paper, we propose a Joint Collaborative Autoencoder framework that learns both user-user and item-item correlations simultaneously, leading to a more robust model and improved top-K recommendation performance. More specifically, we show how to model these user-item correlations and demonstrate the importance of careful normalization to alleviate the influence of feedback heterogeneity. Further, we adopt a pairwise hinge-based objective function to maximize the top-K precision and recall directly for top-K recommenders. Finally, a mini-batch optimization algorithm is proposed to train the proposed model. Extensive experiments on three public datasets show the effectiveness of the proposed framework over state-of-the-art non-neural and neural alternatives.

Tensor completion is a problem of filling the missing or unobserved entries of partially observed tensors. Due to the multidimensional character of tensors in describing complex datasets, tensor completion algorithms and their applications have received wide attention and achievement in areas like data mining, computer vision, signal processing, and neuroscience. In this survey, we provide a modern overview of recent advances in tensor completion algorithms from the perspective of big data analytics characterized by diverse variety, large volume, and high velocity. We characterize these advances from the following four perspectives: general tensor completion algorithms, tensor completion with auxiliary information (variety), scalable tensor completion algorithms (volume), and dynamic tensor completion algorithms (velocity). Further, we identify several tensor completion applications on real-world data-driven problems and present some common experimental frameworks popularized in the literature along with several available software repositories. Our goal is to summarize these popular methods and introduce them to researchers and practitioners for promoting future research and applications. We conclude with a discussion of key challenges and promising research directions in this community for future exploration.

In this paper, we investigate the impact of spatial variation on the construction of location-sensitive user profiles. We demonstrate evidence of spatial variation over a collection of Twitter Lists, wherein we find that crowdsourced labels are constrained by distance. For example, that energy in San Francisco is more associated with the green movement, whereas in Houston it is more associated with oil and gas. We propose a three-step framework for location-sensitive user profiling: first, it constructs a crowdsourced label similarity graph, where each labeler and labelee are annotated with a geographic coordinate; second, it transforms this similarity graph into a directed weighted tree that imposes a hierarchical structure over these labels; third, it embeds this location-sensitive folksonomy into a user profile ranking algorithm that outputs a ranked list of candidate labels for a partially observed user profile. Through extensive experiments over a Twitter list dataset, we demonstrate the effectiveness of this location-sensitive user profiling.

Community detection in complex networks is a fundamental problem that attracts much attention across various disciplines. Previous studies have been mostly focusing on external connections between nodes (i.e., topology structure) in the network whereas largely ignoring internal intricacies (i.e., local behavior) of each node. A pair of nodes without any interaction can still share similar internal behaviors. For example, in an enterprise information network, compromised computers controlled by the same intruder often demonstrate similar abnormal behaviors even if they do not connect with each other. In this paper, we study the problem of community detection in enterprise information networks, where large-scale internal events and external events coexist on each host. The discovered host communities, capturing behavioral affinity, can benefit many comparative analysis tasks such as host anomaly assessment. In particular, we propose a novel community detection framework to identify behavior-based host communities in enterprise information networks, purely based on large-scale heterogeneous event data. We continue proposing an efficient method for assessing host's anomaly level by leveraging the detected host communities. Experimental results on enterprise networks demonstrate the effectiveness of our model.

Tensor-based methods have shown promise in improving upon traditional matrix factorization methods for recommender systems. But tensors may achieve improved recommendation quality while worsening the fairness of the recommendations. Hence, we propose a novel fairness-aware tensor recommendation framework that is designed to maintain quality while dramatically improving fairness. Four key aspects of the proposed framework are: (i) a new sensitive latent factor matrix for isolating sensitive features; (ii) a sensitive information regularizer that extracts sensitive information which can taint other latent factors; (iii) an effective algorithm to solve the proposed optimization model; and (iv) extension to multi-feature and multi-category cases which previous efforts have not addressed. Extensive experiments on real-world and synthetic datasets show that the framework enhances recommendation fairness while preserving recommendation quality in comparison with state-of-the-art alternatives.

How can we efficiently recover missing values for very large-scale real-world datasets that are multi-dimensional even when the auxiliary information is regularized at certain mode? Tensor completion is a useful tool to recover a low-rank tensor that best approximates partially observed data and further predicts the unobserved data by this low-rank tensor, which has been successfully used for many applications such as location-based recommender systems, link prediction, targeted advertising, social media search, and event detection. Due to the curse of dimensionality, existing algorithms for tensor completion that integrate auxiliary information do not scale for tensors with billions of elements. In this paper, we propose DisTenC, a new distributed large-scale tensor completion algorithm that can be distributed on Spark. Our key insights are to (i) efficiently handle trace-based regularization terms; (ii) update factor matrices with caching; and (iii) optimize the update of the new tensor via residuals. In this way, we can tackle the high computational costs of traditional approaches and minimize intermediate data, leading to order-of-magnitude improvements in tensor completion. Experimental results demonstrate that DisTenC is capable of handling up to 10~1000X larger tensors than existing methods with much faster convergence rate, shows better linearity on machine scalability, and achieves up to an average improvement of 23.5% in accuracy in applications.

We propose PsiRec, a novel user preference propagation recommender that incorporates pseudo-implicit feedback for enriching the original sparse implicit feedback dataset. Three of the unique characteristics of PsiRec are: (i) it views user-item interactions as a bipartite graph and models pseudo-implicit feedback from this perspective; (ii) its random walks-based approach extracts graph structure information from this bipartite graph, toward estimating pseudo-implicit feedback; and (iii) it adopts a Skip-gram inspired measure of confidence in pseudo-implicit feedback that captures the pointwise mutual information between users and items. This pseudo-implicit feedback is ultimately incorporated into a new latent factor model to estimate user preference in cases of extreme sparsity. PsiRec results in improvements of 21.5% and 22.7% in terms of Precision@10 and Recall@10 over state-of-the-art Collaborative Denoising Auto-Encoders. Our implementation is available at https://github.com/heyunh2015/PsiRecICDM2018.

Complementary item recommendation finds products that go well with one another (e.g., a camera and a specific lens). While complementary items are ubiquitous, the dimensions by which items go together can vary by both product and category, making it difficult to detect complementary items at scale. Moreover, in practice, user preferences for complementary items can be complex combinations of item quality and evidence of complementarity. Hence, we propose a new neural complementary recommender Encore that can jointly learn complementary item relationships and user preferences. Specifically, Encore (i) effectively combines and balances both stylistic and functional evidence of complementary items across item categories; (ii) naturally models item latent quality for complementary items through Bayesian inference of customer ratings; and (iii) builds a novel neural network model to learn the complex (non-linear) relationships between items for flexible and scalable complementary product recommendations. Through experiments over large Amazon datasets, we find that Encore effectively learns complementary item relationships, leading to an improvement in accuracy of 15.5% on average versus the next-best alternative.

Understanding user interests and expertise is a vital component toward creating rich user models for information personalization in social media, recommender systems and web search. To capture the pair-wise interactions between geo-location and user's topical profile in social-spatial systems, we propose the modeling of fine-grained and multi-dimensional user geo-topic profiles. We then propose a two-layered Bayesian hierarchical user factorization generative framework to overcome user heterogeneity and another enhanced model integrated with user's contextual information to alleviate multi-dimensional sparsity. Through extensive experiments, we find the proposed model leads to a 5\textasciitilde13% improvement in precision and recall over the alternative baselines and an additional 6\textasciitilde11% improvement with the integration of user's contexts.

We propose a system called TwoFace to uncover crowdsourced review manipulators who target online review systems. A unique feature of TwoFace is its three-phase framework:(i) in the first phase, we intelligently sample actual evidence of manipulation(e.g., review manipulators) by exploiting low moderation crowdsourcing platforms that reveal evidence of strategic manipulation;(ii) we then propagate the suspiciousness of these seed users to identify similar users through a random walk over a "suspiciousness»» graph; and(iii) finally, we uncover(hidden) distant users who serve structurally similar roles by mapping users into a low-dimensional embedding space that captures community structure. Altogether, the TwoFace system recovers 83% to 93% of all manipulators in a sample from Amazon of 38,590 reviewers, even when the system is seeded with only a few samples from malicious crowdsourcing sites.

We propose a new model toward improving the quality of image recommendations in social sharing communities like Pinterest, Flickr, and Instagram. Concretely, we propose Neural Personalized Ranking (NPR) -- a personalized pairwise ranking model over implicit feedback datasets -- that is inspired by Bayesian Personalized Ranking (BPR) and recent advances in neural networks. We further build an enhanced model by augmenting the basic NPR model with multiple contextual preference clues including user tags, geographic features, and visual factors. In our experiments over the Flickr YFCC100M dataset, we demonstrate the proposed NPR model is more effective than multiple baselines. Moreover, the contextual enhanced NPR model significantly outperforms the base model by 16.6% and a contextual enhanced BPR model by 4.5% in precision and recall.

research-article Share on ACM Transactions on Interactive Intelligent Systems (TiiS) Special Issue on Trust and Influence in Intelligent Human-Machine Interaction Authors: Benjamin A. Knott The Office of Naval Research Global, Roppongi, Tokyo, Japan The Office of Naval Research Global, Roppongi, Tokyo, JapanView Profile , Jonathan Gratch University of Southern California, USA University of Southern California, USAView Profile , Angelo Cangelosi Plymouth University, USA Plymouth University, USAView Profile , James Caverlee Texas A8M University, USA Texas A8M University, USAView Profile Authors Info & Claims ACM Transactions on Interactive Intelligent SystemsVolume 8Issue 4December 2018 Article No.: 25pp 1–3https://doi.org/10.1145/3281451Published:16 November 2018Publication History 0citation395DownloadsMetricsTotal Citations0Total Downloads395Last 12 Months62Last 6 weeks1 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

I am very happy to report that TSC has gained an Impact Factor (IF) of 3.520 and the 5-year IF of 4.245, both of which represent significant increases from the previous years. This further speaks to the global reputation of the journal and the amazing work done by the past EICs, all the current and past EB members, and reviewers - all of whom have volunteered their precious time despite their very busy schedule to support and contribute to the growth of this journal. I hope to count on your continued engagement for the future growth of this journal. Over this past year, several esteemed EB members have completed their terms of service to TSC after serving for several years. On behalf of the Services Computing community and the TSC EAB, I would like to thank the following Associate Editors who retired from TSC EB in 2017 for their invaluable service and contributions to the journal. Overall, I am very proud of the success that TSC has achieved in 2017. This would not have been possible without the continued support of the authors, readers, reviewers, TSC EAB, TSC EB, and the staff of IEEE and IEEE Computer Society. I look forward to exploring ways to further enhance the reputation and impact of our journal. I would love to hear your suggestions and comments, and I hope to have your continued support.

Crowdsourcing sites heavily rely on paid workers to ensure completion of tasks. Yet, designing a pricing strategies able to incentivize users' quality and retention is non trivial. Existing payment strategies either simply set a fixed payment per task without considering changes in workers' behaviors, or rule out poor quality responses and workers based on coarse criteria. Hence, task requesters may be investing significantly in work that is inaccurate or even misleading. In this paper, we design a dynamic contract to incentivize high-quality work. Our proposed approach offers a theoretically proven algorithm to calculate the contract for each worker in a cost-efficient manner. In contrast to existing work, our contract design is not only adaptive to changes in workers' behavior, but also adjusts pricing policy in the presence of malicious behavior. Both theoretical and experimental analysis over real Amazon review traces show that our contract design can achieve a near optimal solution. Furthermore, experimental results demonstrate that our contract design 1) can promote high-quality work and prevent malicious behavior, and 2) outperforms the intuitive strategy of excluding all malicious workers in terms of the requester's utility.

Tensor completion has become an effective computational tool in many real-world data-driven applications. Beyond traditional static setting, with the increasing popularity of high velocity streaming data, it requires efficient online processing without reconstructing the whole model from scratch. Existing work on streaming tensor completion is usually built upon the assumption that tensors only grow in one mode. Unfortunately, the assumption does not hold in many real-world situations in which tensors may grow in multiple modes, i.e., multi-aspect streaming tensors. Efficiently modeling and completing these incremental tensors without sacrificing its effectiveness remains a challenging task due to the uncertainty of tensor mode changes and complex data structure of multi-aspect streaming tensors. To bridge this gap, we propose a Multi-Aspect Streaming Tensor completion framework (MAST) based on CANDECOMP/PARAFAC (CP) decomposition to track the subspace of general incremental tensors for completion. In addition, we investigate a special situation where time is one mode of the tensors, and leverage its extra structure information to improve the general framework towards higher effectiveness. Experimental results on four datasets collected from various real-world applications demonstrate the effectiveness and efficiency of the proposed framework.

User interests and expertise are valuable but often hidden resources on social media. For example, Twitter Lists and LinkedIn's Skill Tags provide a partial perspective on what users are known for (by aggregating crowd tagging knowledge), but the vast majority of users are untagged; their interests and expertise are essentially hidden from important applications such as personalized recommendation, community detection, and expert mining. A natural approach to overcome these limitations is to intelligently learn user topical profiles by exploiting information from multiple, heterogeneous footprints: for instance, Twitter users who post similar hashtags may have similar interests, and YouTube users who upvote the same videos may have similar preferences. And yet identifying "similar" users by exploiting similarity in such a footprint space often provides conflicting evidence, leading to poor-quality user profiles. In this paper, we propose a unified model for learning user topical profiles that simultaneously considers multiple footprints. We show how these footprints can be embedded in a generalized optimization framework that takes into account pairwise relations among all footprints for robustly learning user profiles. Through extensive experiments, we find the proposed model is capable of learning high-quality user topical profiles, and leads to a 10-15% improvement in precision and mean average error versus a cross-triadic factorization state-of-the-art baseline.

With the rapid adoption of smartphones worldwide and the reliance on app marketplaces to discover new apps, these marketplaces are critical for connecting users with apps. And yet, the user reviews and ratings on these marketplaces may be strategically targeted by app developers. We investigate the use of crowdsourcing platforms to manipulate app reviews. We find that (i) apps targeted by crowdsourcing platforms are rated significantly higher on average than other apps; (ii) the reviews themselves arrive in bursts; (iii) app reviewers tend to repeat themselves by relying on some standard repeated text; and (iv) apps by the same developer tend to share a more similar language model: if one app has been targeted, it is likely that many of the other apps from the same developer have also been targeted.

In this paper, we explore the potential of geo-social media to construct location-based interest profiles to uncover the hidden relationships among disparate locations. Through an investigation of millions of geo-tagged Tweets, we construct a per-city interest model based on fourteen high-level categories (e.g., technology, art, sports). These interest models support the discovery of related locations that are connected based on these categorical perspectives (e.g., college towns or vacation spots) but perhaps not on the individual tweet level. We then connect these city-based interest models to underlying demographic data. By building multivariate multiple linear regression (MMLR) and neural network (NN) models we show how a location's interest profile may be estimated based purely on its demographics features.

This paper tackles the geospatial tag estimation problem, which is of critical importance for location-based search, retrieval, and mining applications. However, tag estimation is challenging due to massive sparsity, uncertainty in the tags actually used, as well as diversity across locations and times. Toward overcoming these challenges, we propose a community-based smoothing approach that seeks to uncover hidden conceptual communities which link multiple related locations by their common interests in addition to their proximity. Through extensive experiments over a sample of millions of geotagged Twitter posts, we demonstrate the effectiveness of the smoothing approach and validate the intuition that geo-locations have the tendency to share similar “ideas” in the formation of conceptual communities.

Modeling, understanding, and predicting the spatio-temporal dynamics of online memes are important tasks, with ramifications on location-based services, social media search, targeted advertising and content delivery networks. However, the raw data revealing these dynamics are often incomplete and error-prone; for example, API limitations and data sampling policies can lead to an incomplete (and often biased) perspective on these dynamics. Hence, in this paper, we investigate new methods for uncovering the full (underlying) distribution through a novel spatio-temporal dynamics recovery framework which models the latent relationships among locations, memes, and times. By integrating these hidden relationships into a tensor-based recovery framework -- called AirCP -- we find that high-quality models of meme spread can be built with access to only a fraction of the full data. Experimental results on both synthetic and real-world Twitter hashtag data demonstrate the promising performance of the proposed framework: an average improvement of over 27% in recovering the spatio-temporal dynamics of hashtags versus five state-of-the-art alternatives.

Crowdsourcing allows many people to complete tasks of various difficulty with minimal recruitment and administration costs. However, the lack of participant accountability may entice people to complete as many tasks as possible without fully engaging in them, jeopardizing the quality of responses. In this paper, we present a dynamic and time efficient solution to the task assignment problem in crowdsourcing platforms. Our proposed approach, CrowdSelect, offers a theoretically proven algorithm to assign workers to tasks in a cost efficient manner, while ensuring high accuracy of the overall task. In contrast to existing works, our approach makes minimal assumptions on the probability of error for workers, and completely removes the assumptions that such probability is known apriori and that it remains consistent over time. Through experiments over real Amazon Mechanical Turk traces and synthetic data, we find that CrowdSelect has a significant gain in term of accuracy compared to state-of-the-art algorithms, and can provide a 17.5\% gain in answers' accuracy compared to previous methods, even when there are over 50\% malicious workers.

We address the challenge of personalized recommendation of high quality content producers in social media. While some candidates are easily identifiable (say, by being "favorited" many times), there is a long-tail of potential candidates for whom we have little evidence. Through careful modeling of contextual factors like the geo-spatial, topical, and social preferences of users, we propose a tensor-based personalized expert recommendation framework that integrates these factors for revealing latent connections between homogeneous entities (e.g., users and users) and between heterogeneous entities (e.g., users and experts). Through extensive experiments over geo-tagged Twitter data, we find that the proposed framework can improve the quality of recommendation by over 30% in both precision and recall compared to the state-of-the-art.

Discovering what people are known for is valuable to many important applications such as recommender systems. Unlike an individual's personal interests, what a user is known for is reflected by the views of others, and is often not easily discerned for a long-tail of the vast majority of users. In this paper, we tackle the problem of discovering what users are known for through a probabilistic model called Bayesian Contextual Poisson Factorization. Moving beyond just modeling user's content, it naturally models and integrates additional contextual factors, concretely, user's geo-spatial footprints and social influence, to overcome noisy online activities and social relations. Through GPS-tagged social media datasets, we find that the proposed method can improve known-for prediction performance by 17.5% in precision and 20.9% in recall on average, and that it can capture the implicit relationships between a user's known-for profile and her content, geo-spatial and social influence.

URL sharing has become one of the most popular activities on many online social media platforms. Shared URLs are an avenue to interesting news articles, memes, photos, as well as low-quality content like spam, promotional ads, and phishing sites. While some URL sharing is organic, other sharing is strategically organized with a common purpose (e.g., aggressively promoting a website). In this paper, we investigate the individual-based and group-based user behavior of URL sharing in social media toward uncovering these organic versus organized user groups. Concretely, we pro- pose a four-phase approach to model, identify, characterize, and classify organic and organized groups who engage in URL sharing. The key motivating insights of this approach are (i) that patterns of individual-based behavioral signals embedded in URL posting activities can uncover groups whose members engage in similar behaviors; and (ii) that group-level behavioral signals can distinguish between organic and organized user groups. Through extensive experiments, we find that levels of organized behavior vary by URL type and that the proposed approach achieves good performance -- an F-measure of 0.836 and Area Under the Curve of 0.921.

We propose a lightweight system for (i) semi-automatically discovering and tracking bias themes associated with opposing sides of a topic; (ii) identifying strong partisans who drive the online discussion; and (iii) inferring the opinion bias of "regular" participants. By taking just two hand-picked seeds to characterize the topic-space (e.g., "pro-choice" and "pro-life") as weak labels, we develop an efficient optimization-based opinion bias propagation method over the social/information network. We show how this approach leads to a 20% accuracy improvement versus a next-best alternative for bias estimation, as well as uncovering the opinion leaders and evolving themes associated with these topics. We also demonstrate how the inferred opinion bias can be integrated into user recommendation, leading to a 26% improvement in precision.

Experts are important for providing reliable and authoritative information and opinion, as well as for improving online reviews and services. While considerable previous research has focused on finding topical experts with broad appeal -- e.g., top Java developers, best lawyers in Texas -- we tackle the problem of personalized expert recommendation, to identify experts who have special personal appeal and importance to users. One of the key insights motivating our approach is to leverage the geo-spatial preferences of users and the variation of these preferences across different regions, topics, and social communities. Through a fine-grained GPS-tagged social media trace, we characterize these geo-spatial preferences for personalized experts, and integrate these preferences into a matrix factorization-based personalized expert recommender. Through extensive experiments, we find that the proposed approach can improve the quality of recommendation by 24% in precision compared to several baselines. We also find that users' geo-spatial preference of expertise and their underlying social communities can ameliorate the cold start problem by more than 20% in precision and recall.

In this workshop we aim to bring together practitioners and researchers to discuss their recent breakthroughs and the challenges with addressing spatial and temporal information access, both from the algorithmic and the architectural perspectives.

Online reviews are a cornerstone of consumer decision making. However, their authenticity and quality has proven hard to control, especially as polluters target these reviews toward promoting products or in degrading competitors. In a troubling direction, the widespread growth of crowdsourcing platforms like Mechanical Turk has created a large-scale, potentially difficult-to-detect workforce of malicious review writers. Hence, this paper tackles the challenge of uncovering crowdsourced manipulation of online reviews through a three-part effort: (i) First, we propose a novel sampling method for identifying products that have been targeted for manipulation and a seed set of deceptive reviewers who have been enlisted through crowdsourcing platforms. (ii) Second, we augment this base set of deceptive reviewers through a reviewer-reviewer graph clustering approach based on a Markov Random Field where we define individual potentials (of single reviewers) and pair potentials (between two reviewers). (iii) Finally, we embed the results of this probabilistic model into a classification framework for detecting crowd-manipulated reviews. We find that the proposed approach achieves up to 0.96 AUC, outperforming both traditional detection methods and a SimRank-based alternative clustering approach.

Social media systems like Twitter and Facebook provide a global infrastructure for sharing information, and in one popular direction, of sharing web hyperlinks. Understanding the behavioral signals of both how URLs are inserted into these systems (via posting by users) and how URLs are received by social media users (via clicking) can provide new insights into social media search, recommendation, and user profiling, among many others. Such studies, however, have traditionally been difficult due to the proprietary (and sometimes private) nature of much URL-related data. Hence, in this paper, we begin a behavioral examination of URL sharing through two distinct perspectives: (i) the first is via a study of how these links are posted through publicly-accessible Twitter data; (ii) the second is via a study of how these links are received by measuring their click patterns through the publicly-accessible Bitly click API. We examine the differences between posting and click patterns in a sample application domain: the classification of spam URLs. We find that these behavioral signals - posting versus clicking - provide overlapping but fundamentally different perspectives on URLs, and that these perspectives can inform the design of future applications of spam link detection and link sharing.

Massive amounts of data are being generated on social media sites, such as Twitter and Facebook. These data can be used to better understand people (e.g., personality traits, perceptions, and preferences) and predict their behavior. As a result, a deeper understanding of users and their behavior can benefit a wide range of intelligent applications, such as advertising, social recommender systems, and personalized knowledge management. These applications will also benefit individual users themselves and optimize their experience across a wide variety of domains, such as retail, healthcare, and education. Since mining and understanding user behavior from social media often requires interdisciplinary effort, including machine learning, text mining, human-computer interaction, and social science, our workshop aims to bring together researchers and practitioners from multiple fields to discuss the creation of deeper models of individual users by mining the content that they publish and the social networking behavior that they exhibit.

This paper addresses the problem of identifying local experts in social media systems like Twitter. Local experts -- in contrast to general topic experts -- have specialized knowledge focused around a particular location, and are important for many applications including answering local information needs and interacting with community experts. And yet identifying these experts is difficult. Hence in this paper, we propose a geo-spatial-driven approach for identifying local experts that leverages the fine-grained GPS coordinates of millions of Twitter users. We propose a local expertise framework that integrates both users' topical expertise and their local authority. Concretely, we estimate a user's local authority via a novel spatial proximity expertise approach that leverages over 15 million geo-tagged Twitter lists. We estimate a user's topical expertise based on expertise propagation over 600 million geo-tagged social connections on Twitter. We evaluate the proposed approach across 56 queries coupled with over 11,000 individual judgments from Amazon Mechanical Turk. We find significant improvement over both general (non-local) expert approaches and comparable local expert finding approaches.

In this paper, we tackle the problem of predicting what online memes will be popular in what locations. Specifically, we develop data-driven approaches building on the global footprint of 755 million geo-tagged hashtags spread via Twitter. Our proposed methods model the geo-spatial propagation of online information spread to identify which hashtags will become popular in specific locations. Concretely, we develop a novel reinforcement learning approach that incrementally updates the best geo-spatial model. In experiments, we find that the proposed method outperforms alternative linear regression based methods.

Massive amounts of data are being generated on social media sites, such as Twitter and Facebook. These data can be used to better understand people (e.g., personality traits, perceptions, and preferences) and predict their behavior. As a result, a deeper understanding of users and their behavior can benefit a wide range of intelligent applications, such as advertising, social recommender systems, and personalized knowledge management. These applications will also benefit individual users themselves and optimize their experience across a wide variety of domains, such as retail, healthcare, and education. Since mining and understanding user behavior from social media often requires interdisciplinary effort, including machine learning, text mining, human-computer interaction, and social science, our workshop aims to bring together researchers and practitioners from multiple fields to discuss the creation of deeper models of individual users by mining the content that they publish and the social networking behavior that they exhibit.

We propose a novel network-based approach for location estimation in social media that integrates evidence of the social tie strength between users for improved location estimation. Concretely, we propose a location estimator -- FriendlyLocation -- that leverages the relationship between the strength of the tie between a pair of users, and the distance between the pair. Based on an examination of over 100 million geo-encoded tweets and 73 million Twitter user profiles, we identify several factors such as the number of followers and how the users interact that can strongly reveal the distance between a pair of users. We use these factors to train a decision tree to distinguish between pairs of users who are likely to live nearby and pairs of users who are likely to live in different areas. We use the results of this decision tree as the input to a maximum likelihood estimator to predict a user's location. We find that this proposed method significantly improves the results of location estimation relative to a state-of-the-art technique. Our system reduces the average error distance for 80% of Twitter users from 40 miles to 21 miles using only information from the user's friends and friends-of-friends, which has great significance for augmenting traditional social media and enriching location-based services with more refined and accurate location estimates.

We conduct a study of the spatio-temporal dynamics of Twitter hashtags through a sample of 2 billion geo-tagged tweets. In our analysis, we (i) examine the impact of location, time, and distance on the adoption of hashtags, which is important for understanding meme diffusion and information propagation; (ii) examine the spatial propagation of hashtags through their focus, entropy, and spread; and (iii) present two methods that leverage the spatio-temporal propagation of hashtags to characterize locations. Based on this study, we find that although hashtags are a global phenomenon, the physical distance between locations is a strong constraint on the adoption of hashtags, both in terms of the hashtags shared between locations and in the timing of when these hashtags are adopted. We find both spatial and temporal locality as most hashtags spread over small geographical areas but at high speeds. We also find that hashtags are mostly a local phenomenon with long-tailed life spans. These (and other) findings have important implications for a variety of systems and applications, including targeted advertising, location-based services, social media search, and content delivery networks.

Highly dynamic real-time microblog systems have already published petabytes of real-time human sensor data in the form of status updates. However, the lack of user adoption of geo-based features per user or per post signals that the promise of microblog services as location-based sensing systems may have only limited reach and impact. Thus, in this article, we propose and evaluate a probabilistic framework for estimating a microblog user's location based purely on the content of the user's posts. Our framework can overcome the sparsity of geo-enabled features in these services and bring augmented scope and breadth to emerging location-based personalized information services. Three of the key features of the proposed approach are: (i) its reliance purely on publicly available content; (ii) a classification component for automatically identifying words in posts with a strong local geo-scope; and (iii) a lattice-based neighborhood smoothing model for refining a user's location estimate. On average we find that the location estimates converge quickly, placing 51% of users within 100 miles of their actual location.

In this manuscript, we study the problem of detecting coordinated free text campaigns in large-scale social media. These campaigns—ranging from coordinated spam messages to promotional and advertising campaigns to political astro-turfing—are growing in significance and reach with the commensurate rise in massive-scale social systems. Specifically, we propose and evaluate a content-driven framework for effectively linking free text posts with common “talking points” and extracting campaigns from large-scale social media. Three of the salient features of the campaign extraction framework are: (i) first, we investigate graph mining techniques for isolating coherent campaigns from large message-based graphs; (ii) second, we conduct a comprehensive comparative study of text-based message correlation in message and user levels; and (iii) finally, we analyze temporal behaviors of various campaign types. Through an experimental study over millions of Twitter messages we identify five major types of campaigns—namely Spam, Promotion, Template, News, and Celebrity campaigns—and we show how these campaigns may be extracted with high precision and recall.

In this paper, we propose and evaluate a novel content-driven crowd discovery algorithm that can efficiently identify newly-formed communities of users from the real-time web. Short-lived crowds reflect the real-time interests of their constituents and provide a foundation for user-focused web monitoring. Three of the salient features of the algorithm are its: (i) prefix-tree based locality-sensitive hashing approach for discovering crowds from high-volume rapidly-evolving social media; (ii) efficient user profile updating for incorporating new user activities and fading older ones; and (iii) key dimension identification, so that crowd detection can be focused on the most active portions of the real-time web. Through extensive experimental study, we find significantly more efficient crowd discovery as compared to both a k-means clustering-based approach and a MapReduce-based implementation, while maintaining high-quality crowds as compared to an offline approach. Additionally, we find that expert crowds tend to be "stickier" and last longer in comparison to crowds of typical users.

In this paper we seek to understand and model the global spread of social media. How does social media spread from location to location across the globe? Can we model this spread and predict where social media will be popular in the future? Toward answering these questions, we develop a probabilistic model that synthesizes two conflicting hypotheses about the nature of online information spread: (i) the spatial influence model, which asserts that social media spreads to locations that are close by; and (ii) the community affinity influence model, which asserts that social media spreads between locations that are culturally connected, even if they are distant. Based on the geospatial footprint of 755 million geo-tagged hashtags spread through Twitter, we evaluate these models at predicting locations that will adopt hashtags in the future. We find that distance is the single most important explanation of future hashtag adoption since hashtags are fundamentally local. We also find that community affinities (like culture, language, and common interests) enhance the quality of purely spatial models, indicating the necessity of incorporating non-spatial features into models of global social media spread.

Massive amounts of data are being generated on social media sites, such as Twitter and Facebook. This data can be used to better understand people, such as their personality traits, perceptions, and preferences, and predict their behavior. This deeper understanding of users and their behaviors can benefit a wide range of intelligent applications, such as advertising, social recommender systems, and personalized knowledge management. These applications will also benefit individual users themselves by optimizing their experiences across a wide variety of domains, such as retail, healthcare, and education. Since mining and understanding user behavior from social media often requires interdisciplinary effort, including machine learning, text mining, human-computer interaction, and social science, our workshop aims to bring together researchers and practitioners from multiple fields to discuss the creation of deeper models of individual users by mining the content that they publish and the social networking behavior that they exhibit.

Large-scale socially-generated metadata – like user-contributed tags, comments, and ratings – is one of the key features driving the growth and success of the emerging Social Web. While tags and ratings provide succinct metadata about Social Web cont

The emergence of location sharing services is rapidly accelerating the convergence of our online and offline activities. In one direction, Foursquare, Google Latitude, Facebook Places, and related services are enriching real-world venues with the social and semantic connections among online users. In analogy to how clickstreams have been successfully incorporated into traditional web ranking based on content and link analysis, we propose to mine traffic patterns revealed through location sharing services to augment traditional location-based search. Concretely, we study location-based traffic patterns revealed through location sharing services and find that these traffic patterns can identify semantically related locations. Based on this observation, we propose and evaluate a traffic-driven location clustering algorithm that can group semantically related locations with high confidence. Through experimental study of 12 million locations from Foursquare, we extend this result through supervised location categorization, wherein traffic patterns can be used to accurately predict the semantic category of uncategorized locations. Based on these results, we show how traffic-driven semantic organization of locations may be naturally incorporated into location-based web search.

We study the problem of efficient discovery of trending phrases from high-volume text streams -- be they sequences of Twitter messages, email messages, news articles, or other time-stamped text documents. Most existing approaches return top-k trending phrases. But, this approach neither guarantees that the top-k phrases returned are all trending, nor that all trending phrases are returned. In addition, the value of k is difficult to set and is indifferent to stream dynamics. Hence, we propose an approach that identifies all the trending phrases in a stream and is flexible to the changing stream properties.

We study the problem of detecting coordinated free text campaigns in large-scale social media. These campaigns -- ranging from coordinated spam messages to promotional and advertising campaigns to political astro-turfing -- are growing in significance and reach with the commensurate rise of massive-scale social systems. Often linked by common "talking points", there has been little research in detecting these campaigns. Hence, we propose and evaluate a content-driven framework for effectively linking free text posts with common "talking points" and extracting campaigns from large-scale social media. One of the salient aspects of the framework is an investigation of graph mining techniques for isolating coherent campaigns from large message-based graphs. Through an experimental study over millions of Twitter messages we identify five major types of campaigns -- Spam, Promotion, Template, News, and Celebrity campaigns -- and we show how these campaigns may be extracted with high precision and recall.

In this paper, we investigate the interplay of distance and tie strength through an examination of 20 million geo-encoded tweets collected from Twitter and 6 million user profiles. Concretely, we investigate the relationship between the strength of the tie between a pair of users, and the distance between the pair. We identify several factors -- including following, mentioning, and actively engaging in conversations with another user -- that can strongly reveal the distance between a pair of users. We find a bimodal distribution in Twitter, with one peak around 10 miles from people who live nearby, and another peak around 2500 miles, further validating Twitter's use as both a social network (with geographically nearby friends) and as a news distribution network (with very distant relationships).

CrowdTracker is a community-based web monitoring system optimized for real-time web streams like Twitter, Facebook, and Google Buzz. In this demo summary, we provide an overview of the system and architecture, and outline the demonstration plan.

In this paper, we study the problem of automatically discovering and tracking transient crowds in highly-dynamic social messaging systems like Twitter and Facebook. Unlike the more static and long-lived group-based membership offered on many social networks (e.g., fan of the LA Lakers), a transient crowd is a short-lived ad-hoc collection of users, representing a "hotspot" on the real-time web. Successful detection of these hotspots can positively impact related research directions in online event detection, content personalization, social information discovery, etc. Concretely, we propose to model crowd formation and dispersion through a message-based communication clustering approach over time-evolving graphs that captures the natural conversational nature of social messaging systems. Two of the salient features of the proposed approach are (i) an efficient locality- based clustering approach for identifying crowds of users in near real-time compared to more heavyweight static clustering algorithms; and (ii) a novel crowd tracking and evolution approach for linking crowds across time periods. We find that the locality-based clustering approach results in empirically high-quality clusters relative to static graph clus- tering techniques at a fraction of the computational cost. Based on a three month snapshot of Twitter consisting of 711,612 users and 61.3 million messages, we show how the proposed approach can successfully identify and track interesting crowds based on the Twitter communication structure and uncover crowd-based topics of interest.

We propose and evaluate a probabilistic framework for estimating a Twitter user's city-level location based purely on the content of the user's tweets, even in the absence of any other geospatial cues. By augmenting the massive human-powered sensing capabilities of Twitter and related microblogging services with content-derived location information, this framework can overcome the sparsity of geo-enabled features in these services and enable new location-based personalized information services, the targeting of regional advertisements, and so on. Three of the key features of the proposed approach are: (i) its reliance purely on tweet content, meaning no need for user IP information, private login information, or external knowledge bases; (ii) a classification component for automatically identifying words in tweets with a strong local geo-scope; and (iii) a lattice-based neighborhood smoothing model for refining a user's location estimate. The system estimates k possible locations for each user in descending order of confidence. On average we find that the location estimates converge quickly (needing just 100s of tweets), placing 51% of Twitter users within 100 miles of their actual location.

We study the problem of automatically identifying ``hotspots'' on the real-time web. Concretely, we propose to identify highly-dynamic ad-hoc collections of users -- what we refer to as crowds -- in massive social messaging systems like Twitter and Facebook. The proposed approach relies on a message-based communication clustering approach over time-evolving graphs that captures the natural conversational nature of social messaging systems. One of the salient features of the proposed approach is an efficient locality-based clustering approach for identifying crowds of users in near real-time compared to more heavyweight static clustering algorithms. Based on a three month snapshot of Twitter consisting of 711,612 users and 61.3 million messages, we show how the proposed approach can efficiently and effectively identify Twitter-based crowds relative to static graph clustering techniques at a fraction of the computational cost.

Social information systems - popularized by Facebook, Wikipedia, Twitter, and other social websites - are emerging as a powerful new paradigm for distributed social-powered information management. While there has been growing interest in these systems by businesses, government agencies, and universi

Web-based social systems enable new community-based opportunities for participants to engage, share, and interact. This community value and related services like search and advertising are threatened by spammers, content polluters, and malware disseminators. In an effort to preserve community value and ensure longterm success, we propose and evaluate a honeypot-based approach for uncovering social spammers in online social systems. Two of the key components of the proposed approach are: (1) The deployment of social honeypots for harvesting deceptive spam profiles from social networking communities; and (2) Statistical analysis of the properties of these spam profiles for creating spam classifiers to actively filter out existing and new spammers. We describe the conceptual framework and design considerations of the proposed approach, and we present concrete observations from the deployment of social honeypots in MySpace and Twitter. We find that the deployed social honeypots identify social spammers with low false positive rates and that the harvested spam data contains signals that are strongly correlated with observable profile features (e.g., content, friend information, posting patterns, etc.). Based on these profile features, we develop machine learning based classifiers for identifying previously unknown spammers with high precision and a low rate of false positives.

We present the conceptual framework of the Social Honeypot Project for uncovering social spammers who target online communities and initial empirical results from Twitter and MySpace. Two of the key components of the Social Honeypot Project are: (1) The deployment of social honeypots for harvesting deceptive spam profiles from social networking communities; and (2) Statistical analysis of the properties of these spam profiles for creating spam classifiers to actively filter out existing and new spammers.

Abstract This paper studies how varied damping factors in the PageRank algorithm influence the ranking of authors and proposes weighted PageRank algorithms. We selected the 108 most highly cited authors in the information retrieval (IR) area from the 1970s to 2008 to form the author co‐citation network. We calculated the ranks of these 108 authors based on PageRank with the damping factor ranging from 0.05 to 0.95. In order to test the relationship between different measures, we compared PageRank and weighted PageRank results with the citation ranking, h‐index, and centrality measures. We found that in our author co‐citation network, citation rank is highly correlated with PageRank with different damping factors and also with different weighted PageRank algorithms; citation rank and PageRank are not significantly correlated with centrality measures; and h‐index rank does not significantly correlate with centrality measures but does significantly correlate with other measures. The key factors that have impact on the PageRank of authors in the author co‐citation network are being co‐cited with important authors.

Link-based analysis of the Web provides the basis for many important applications-like Web search, Web-based data mining, and Web page categorization-that bring order to the massive amount of distributed Web content. Due to the overwhelming reliance on these important applications, there is a rise in efforts to manipulate (or spam) the link structure of the Web. In this manuscript, we present a parameterized framework for link analysis of the Web that promotes spam resilience through a source-centric view of the Web. We provide a rigorous study of the set of critical parameters that can impact source-centric link analysis and propose the novel notion of influence throttling for countering the influence of link-based manipulation. Through formal analysis and a large-scale experimental study, we show how different parameter settings may impact the time complexity, stability, and spam resilience of Web link analysis. Concretely, we find that the source-centric model supports more effective and robust rankings in comparison with existing Web algorithms such as PageRank.

Web spam is a widely-recognized threat to the quality and security of the Web. Web spam pages pollute search engine indexes, burden Web crawlers and Web mining services, and expose users to dangerous Web-borne malware. To defend against Web spam, most previous research analyzes the contents of Web pages and the link structure of the Web graph. Unfortunately, these heavyweight approaches require full downloads of both legitimate and spam pages to be effective, making real-time deployment of these techniques infeasible for Web browsers, high-performance Web crawlers, and real-time Web applications. In this paper, we present a lightweight, predictive approach to Web spam classification that relies exclusively on HTTP session information (i.e., hosting IP addresses and HTTP session headers).

One of the cornerstones of the Social Web is informal user-generated metadata (or tags) for annotating web objects like pages, images, and videos. However, many real-world domains are currently left out of the social tagging phenomenon due to the lack of a wide-scale tagging-savvy audience - domains like the personal desktop, enterprise intranets, and digital libraries. Hence in this paper, we propose a lightweight interactive tagging framework for providing high-quality tag suggestions for the vast majority of untagged content. One of the salient features of the proposed framework is its incorporation of user feedback for iteratively refining tag suggestions. Concretely, we describe and evaluate three feedback models - Tag-Based, Term-Based, and Tag Co-location. Through extensive user evaluation and testing, we find that feedback can significantly improve tag quality with minimal user involvement.

We propose the SocialTrust framework for tamper-resilient trust establishment in online social networks. Two of the salient features of SocialTrust are its dynamic revision of trust by (i) distinguishing relationship quality from trust; and (ii) incorporating a personalized feedback mechanism for adapting as the social network evolves.

We present the design of Plurality, an interactive tagging system. Plurality's modular architecture allows users to automatically generate high-quality tags over Web content, as well as over archival and personal content typically beyond the reach of existing Web 2.0 social tagging systems. Three of the salient features of Plurality are: (i) its self-learning and feedback-sensitive capabilities based on a user's personalized tagging style; (ii) its leveraging of the collective intelligence of existing social tagging services; and (iii) its context-awareness for optimizing tag suggestions, e.g., based on spatial or temporal features.

We describe DSphere - a decentralized system for crawling, indexing, searching and ranking of documents in the World Wide Web. Unlike most of the existing search technologies that depend heavily on a page-centric view of the Web, we advocate a source-centric view of the Web and propose a decentralized architecture for crawling, indexing and searching the Web in a distributed source-specific fashion. A fully decentralized crawler is developed to crawl the World Wide Web where each peer is assigned the responsibility of crawling a specific set of documents referred to as a source collection. Link analysis techniques are used for ranking documents. Traditional link analysis techniques suffer from problems like slow refresh rate and vulnerabilities to Web Spam. We propose a source-based link analysis approach, which computes fast and accurate ranking scores for all crawled documents.

Web search is one of the most critical applications for managing the massive amount of distributed Web content. Due to the overwhelming reliance on Web search, there is a rise in efforts to manipulate (or spam) Web search engines. In this paper, we develop a spam-resilient ranking model that promotes a source-based view of the Web. One of the most salient features of our spam-resilient ranking algorithm is the concept of influence throttling. We show how to utilize influence throttling to counter Web spam that aims at manipulating link-based ranking systems, especially PageRank-like systems. Through formal analysis and experimental evaluation, we show the effectiveness and robustness of our spam-resilient ranking model in comparison with existing Web algorithms such as PageRank.

We introduce the concept of link credibility, identify the conflation of page quality and link credibility in popular Web link analysis algorithms, and discuss how to decouple link credibility from page quality. Our credibility-based link analysis exhibits three distinct features. First, we develop several techniques for semi-automatically assessing link credibility for all Web pages. Second, our link credibility assignment algorithms allow users to assess credibility in a personalized manner. Third, we develop a novel credibility-based Web ranking algorithm - CredibleRank - which incorporates credibility information directly into the quality assessment of each page on the Web. Our experimental study shows that our approach is significantly and consistently more spam-resilient than both PageRank and TrustRank.

Web ranking is one of the most successful and widely used collaborative computing applications, in which Web pages collaborate in the form of varying degree of relationships to assess their relative quality. Though many observe that links display strong source-centric locality, for example, in terms of administrative domains and hosts, most Web ranking analysis to date has focused on the flat page-level Web linkage structure. In this paper we develop a framework for link-based collaborative ranking of the Web by utilizing the strong Web link structure. We argue that this source-centric link analysis is promising since it captures the natural link-locality structure of the Web, can provide more appealing and efficient Web applications, and reflects many natural types of structured human collaborations. Concretely, we propose a generic framework for source-centric collaborative ranking of the Web. This paper makes two unique contributions. First, we provide a rigorous study of the set of critical parameters that can impact source-centric link analysis, such as source size, the presence of self-links, and different source-citation link weighting schemes (e.g., uniform, link count, source consensus). Second, we conduct a large-scale experimental study to understand how different parameter settings may impact the time complexity, stability, and spam-resilience of Web ranking. We find that careful tuning of these parameters is vital to ensure success over each objective and to balance the performance across all objectives

Business processes continue to play an important role in today's service-oriented enterprise computing systems. Mining, discovering, and integrating process-oriented services has attracted growing attention in the recent year. In this paper we present a quantitative approach to modeling and capturing the similarity and dissimilarity between different process designs. We derive the similarity measures by analyzing the process dependency graphs of the participating workflow processes. We first convert each process dependency graph into a normalized process matrix. Then we calculate the metric space distance between the normalized matrices. This distance measure can be used as a quantitative and qualitative tool in process mining, process merging, and process clustering, and ultimately it can reduce or minimize the costs involved in design, analysis, and evolution of workflow systems

We present an adaptive distributed query-sampling framework that is quality-conscious for extracting high-quality text database samples. The framework divides the query-based sampling process into an initial seed sampling phase and a quality-aware iterative sampling phase. In the second phase the sampling process is dynamically scheduled based on estimated database size and quality parameters derived during the previous sampling process. The unique characteristic of our adaptive query-based sampling framework is its self-learning and self-configuring ability based on the overall quality of all text databases under consideration. We introduce three quality-conscious sampling schemes for estimating database quality, and our initial results show that the proposed framework supports higher-quality document sampling than existing approaches.

This paper presents DynaBot, a domain-specific Web service discovery system. The core idea of the DynaBot service discovery system is to use domain-specific service class descriptions powered by an intelligent Deep Web crawler. In contrast to current registry-based service discovery systems -like the several available UDDI registries - DynaBot promotes focused crawling of the Deep Web of services and discovers candidate services that are relevant to the domain of interest. It uses intelligent filtering algorithms to match services found by focused crawling with the domain-specific service class descriptions. We demonstrate the capability of DynaBot through the BLAST scenario and describe our initial experience with DynaBot.

This paper presents the QA-Pagelet as a fundamental data preparation technique for large-scale data analysis of the deep Web. To support QA-Pagelet extraction, we present the Thor framework for sampling, locating, and partioning the QA-Pagelets from the deep Web. Two unique features of the Thor framework are 1) the novel page clustering for grouping pages from a deep Web source into distinct clusters of control-flow dependent pages and 2) the novel subtree filtering algorithm that exploits the structural and content similarity at subtree level to identify the QA-Pagelets within highly ranked page clusters. We evaluate the effectiveness of the Thor framework through experiments using both simulation and real data sets. We show that Thor performs well over millions of deep Web pages and over a wide range of sources, including e-commerce sites, general and specialized search engines, corporate Web sites, medical and legal resources, and several others. Our experiments also show that the proposed page clustering algorithm achieves low-entropy clusters, and the subtree filtering algorithm identifies QA-Pagelets with excellent precision and recall.

We introduce the concept of a QA-Pagelet to refer to the content region in a dynamic page that contains query matches. We present THOR, a scalable and efficient mining system for discovering and extracting QA-Pagelets from the deep Web. A unique feature of THOR is its two-phase extraction framework. In the first phase, pages from a deep Web site are grouped into distinct clusters of structurally-similar pages. In the second phase, pages from each page cluster are examined through a subtree filtering algorithm that exploits the structural and content similarity at subtree level to identify the QA-Pagelets.

In this paper we present a personalized web service discovery and ranking technique for discovering and ranking relevant data-intensive web services. Our first prototype -- called BASIL -- supports a personalized view of data-intensive web services through source-biased focus. BASIL provides service discovery and ranking through source-biased probing and source-biased relevance metrics. Concretely, the BASIL approach has three unique features: (1) It is able to determine in very few interactions whether a target service is relevant to the given source service by probing the target with very precise probes; (2) It can evaluate and rank the relevant services discovered based on a set of source-biased relevance metrics; and (3) It can identify interesting types of relationships for each source service with respect to other discovered services, which can be used as value-added metadata for each service. We also introduce a performance optimization technique called source-biased probing with focal terms to further improve the effectiveness of the basic source-biased service discovery algorithm. The paper concludes with a set of initial experiments showing the effectiveness of the BASIL system.