Xiusi Chen, Jyun-Yu Jiang, Wei-Cheng Chang, Cho-Jui Hsieh, Hsiang-Fu Yu, Wei Wang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

With the advent of generative deep learning models, generative IR has gained increasing attention. However, existing methods face two issues: (1) when a document is represented by a single semantic ID, the retrieval model may fail to capture the multifaceted and complex content of the document; and (2) when the generated training data exhibits semantic ambiguity, the retrieval model may struggle to distinguish the differences in the content of similar documents. To address these issues, we propose Multi-DSI to (1) offer multiple non-deterministic semantic identifiers and (2) align the concepts of queries and documents to avoid ambiguity. Extensive experiments on two benchmark datasets demonstrate that Multi-DSI significantly outperforms baseline methods by 7.4%.

Embedding-based Retrieval Models (ERMs) have emerged as a promising framework for large-scale text retrieval problems due to powerful large language models. Nevertheless, fine-tuning ERMs to reach state-of-the-art results can be expensive due to the extreme scale of data as well as the complexity of multi-stages pipelines (e.g., pre-training, fine-tuning, distillation). In this work, we propose the PEFA framework, namely ParamEter-Free Adapters, for fast tuning of ERMs without any backward pass in the optimization. At index building stage, PEFA equips the ERM with a non-parametric k-nearest neighbor (kNN) component. At inference stage, PEFA performs a convex combination of two scoring functions, one from the ERM and the other from the kNN. Based on the neighborhood definition, PEFA framework induces two realizations, namely PEFA-XL (i.e., extra large) using double ANN indices and PEFA-XS (i.e., extra small) using a single ANN index. Empirically, PEFA achieves significant improvement on two retrieval applications. For document retrieval, regarding Recall@100 metric, PEFA improves not only pre-trained ERMs on Trivia-QA by an average of 13.2%, but also fine-tuned ERMs on NQ-320K by an average of 5.5%, respectively. For product search, PEFA improves the Recall@100 of the fine-tuned ERMs by an average of 5.3% and 14.5%, for PEFA-XS and PEFA-XL, respectively. Our code is available at https://github.com/amzn/pecos/tree/mainline/examples/pefa-wsdm24.

Entity disambiguation is one of the most important natural language tasks to identify entities behind ambiguous surface mentions within a knowledge base. Although many recent studies apply deep learning to achieve decent results, they need exhausting pre-training and mediocre recall in the retrieval stage. In this paper, we propose a novel framework, eXtreme Multi-label Ranking for Entity Disambiguation (XMRED), to address this challenge. An efficient zero-shot entity retriever with auxiliary data is first pre-trained to recall relevant entities based on linear models. Specifically, the retrieval process can be considered as an extreme multi-label ranking (XMR) task. Entities are first clustered at different scales to form a label tree, thereby learning multi-scale entity retrievers over the label tree with high recall. Moreover, XMRED applies deep cross-encoder as a re-ranker to achieve high precision based on high-quality candidates. Extensive experimental results based on the AIDA-CoNLL benchmark and five zero-shot testing datasets demonstrate that XMRED obtains 98% and over 95% recall scores for in-domain and zero-shot datasets with top-10 retrieved entities. With a deep cross-encoder as the re-ranker, XMRED further outperforms the previous state-of-the-art by 1.74% in In-KB micro-F1 scores on average with a significant improvement on the training efficiency from days to 3.48 hours. In addition, XMRED also beats the state-of-the-art for page-level document retrieval by 2.38% in accuracy and 1.90% in recall@5.

Nowadays, modern recommender systems usually leverage textual and visual contents as auxiliary information to predict user preference. For textual information, review texts are one of the most popular contents to model user behaviors. Nevertheless, reviews usually lose their shine when it comes to top-N recommender systems because those that solely utilize textual reviews as features struggle to adequately capture the interaction relationships between users and items. For visual one, it is usually modeled with naive convolutional networks and also hard to capture high-order relationships between users and items. Moreover, previous works did not collaboratively use both texts and images in a proper way. In this paper, we propose printf, preference modeling based on user reviews with item images and textual information via graph learning, to address the above challenges. Specifically, the dimension-based attention mechanism directs relations between user reviews and interacted items, allowing each dimension to contribute different importance weights to derive user representations. Extensive experiments are conducted on three publicly available datasets. The experimental results demonstrate that our proposed printf consistently outperforms baseline methods with the relative improvements for NDCG@5 of 26.80%, 48.65%, and 25.74% on Amazon-Grocery, Amazon-Tools, and Amazon-Electronics datasets, respectively. The in-depth analysis also indicates the dimensions of review representations definitely have different topics and aspects, assisting the validity of our model design.

The semantic matching problem in product search seeks to retrieve all semantically relevant products given a user query. Recent studies have shown that extreme multi-label classification~(XMC) model enjoys both low inference latency and high recall in real-world scenarios. These XMC semantic matching models adopt TF-IDF vectorizers to extract query text features and use mainly sparse matrices for the model weights. However, limited availability of libraries for efficient parallel sparse modules may lead to tediously long model building time when the problem scales to hundreds of millions of labels. This incurs significant hardware cost and renders the semantic model stale even before it is deployed. In this paper, we investigate and accelerate the model building procedures in a tree-based XMC model. On a real-world semantic matching task with 100M labels, our enhancements achieve over 10 times acceleration (from 3.1 days to 6.7 hours) while reducing hardware cost by 25%.

The eXtreme Multi-label Classification (XMC) problem seeks to find relevant labels from an exceptionally large label space. Most of the existing XMC learners focus on the extraction of semantic features from input query text. However, conventional XMC studies usually neglect the side information of instances and labels, which can be of use in many real-world applications such as recommendation systems and e-commerce product search. We propose Predicted Instance Neighborhood Aggregation (PINA), a data augmentation method for the general XMC problem that leverages beneficial side information. Unlike most existing XMC frameworks that treat labels and input instances as featureless indicators and independent entries, PINA extracts information from the label metadata and the correlations among training instances. Extensive experimental results demonstrate the consistent gain of PINA on various XMC tasks compared to the state-of-the-art methods: PINA offers a gain in accuracy compared to standard XR-Transformers on five public benchmark datasets. Moreover, PINA achieves a $\sim 5$% gain in accuracy on the largest dataset LF-AmazonTitles-1.3M.

As influencers play considerable roles in social media marketing, companies increase the budget for influencer marketing. Hiring effective influencers is crucial in social influencer marketing, but it is challenging to find the right influencers among hundreds of millions of social media users. In this paper, we propose InfluencerRank that ranks influencers by their effectiveness based on their posting behaviors and social relations over time. To represent the posting behaviors and social relations, the graph convolutional neural networks are applied to model influencers with heterogeneous networks during different historical periods. By learning the network structure with the embedded node features, InfluencerRank can derive informative representations for influencers at each period. An attentive recurrent neural network finally distinguishes highly effective influencers from other influencers by capturing the knowledge of the dynamics of influencer representations over time. Extensive experiments have been conducted on an Instagram dataset that consists of 18,397 influencers with their 2,952,075 posts published within 12 months. The experimental results demonstrate that InfluencerRank outperforms existing baseline methods. An in-depth analysis further reveals that all of our proposed features and model components are beneficial to discover effective influencers.

Few-shot question answering (QA) aims at precisely discovering answers to a set of questions from context passages while only a few training samples are available. Although existing studies have made some progress and can usually achieve proper results, they suffer from understanding deep semantics for reasoning out the questions. In this paper, we develop Gotta, a Generative prOmpT-based da Ta Augmentation framework to mitigate the challenge above. Inspired by the human reasoning process, we propose to integrate the cloze task to enhance few-shot QA learning. Following the recent success of prompt-tuning, we present the cloze task in the same format as the main QA task, allowing the model to learn both tasks seamlessly together to fully take advantage of the power of prompt-tuning. Extensive experiments on widely used benchmarks demonstrate that Gotta consistently outperforms competitive baselines, validating the effectiveness of our proposed prompt-tuning-based cloze task, which not only fine-tunes language models but also learns to guide reasoning in QA tasks. Further analysis shows that the prompt-based loss incorporates the auxiliary task better than the multi-task loss, highlighting the strength of prompt-tuning on the few-shot QA task.

Uncertainty quantification is one of the most crucial tasks to obtain trustworthy and reliable machine learning models for decision making. However, most research in this domain has only focused on problems with small label spaces and ignored eXtreme Multi-label Classification (XMC), which is an essential task in the era of big data for web-scale machine learning applications. Moreover, enormous label spaces could also lead to noisy retrieval results and intractable computational challenges for uncertainty quantification. In this paper, we aim to investigate general uncertainty quantification approaches for tree-based XMC models with a probabilistic ensemble-based framework. In particular, we analyze label-level and instance-level uncertainty in XMC, and propose a general approximation framework based on beam search to efficiently estimate the uncertainty with a theoretical guarantee under long-tail XMC predictions. Empirical studies on six large-scale real-world datasets show that our framework not only outperforms single models in predictive performance, but also can serve as strong uncertainty-based baselines for label misclassification and out-of-distribution detection, with significant speedup. Besides, our framework can further yield better state-of-the-art results based on deep XMC models with uncertainty quantification.

Approximate K-Nearest Neighbor Search (AKNNS) has now become ubiquitous in modern applications, such as a fast search procedure with two-tower deep learning models. Graph-based methods for AKNNS in particular have received great attention due to their superior performance. These methods rely on greedy graph search to traverse the data points as embedding vectors in a database. Under this greedy search scheme, we make a key observation: many distance computations do not influence search updates so that these computations can be approximated without hurting performance. As a result, we propose FINGER, a fast inference method for efficient graph search in AKNNS. FINGER approximates the distance function by estimating angles between neighboring residual vectors. The approximated distance can be used to bypass unnecessary computations for faster searches. Empirically, when it comes to speeding up the inference of HNSW, which is one of the most popular graph-based AKNNS methods, FINGER significantly outperforms existing acceleration approaches and conventional libraries by 20 to 60 across different benchmark datasets.

Professional basketball provides an intriguing example of a dynamic spatio-temporal game that incorporates both hidden strategy policies and situational decision making. During a game, the coaches and players are assumed to follow a general game plan, but players are also forced to make spur-of-the-moment decisions based on immediate conditions on the court. However, because it is challenging to process heterogeneous signals on the court and the space of potential actions and outcomes is massive, it is hard for players to find an optimal strategy on the fly given a short amount of time to observe conditions and take action. In this work, we present ReLiable (ReinforcemEnt Learning In bAsketBaLl gamEs). Specifically, we investigate the possibility of using reinforcement learning (RL) to guide player decisions. We train an offline deep Q-network (DQN) on historical National Basketball Association (NBA) game data from 2015-2016. The data include play-by-play and player movement sensor data. We apply our trained agent to games that it has not seen. Our method is able to propose potentially smarter tactical strategies, compared with replay gameplay data, producing expected final game scores comparable to elite NBA teams. Our approach can be useful for learning strategy policies from other game-like domains characterized by competing groups and sequential spatio-temporal event data.

A massive amount of research on graph representation learning has been carried out to learn dense features as graph embedding for information networks, thereby capturing the semantics in complex networks and benefiting a variety of downstream tasks. Most of the existing studies focus on structural properties, such as distances and neighborhood proximity between nodes. However, real-world information networks are dominated by the low-degree nodes because they are not only sparse but also subject to the Power law form. Due to the sparsity, proximity-based methods are incapable of deriving satisfactory representations for these tail nodes. To address this challenge, we propose a novel approach, Content-Preserving Locality-Sensitive Hashing~(CP-LSH), by incorporating the content information for representation learning. Specifically, we aim at preserving LSH-based content similarity between nodes to leverage the knowledge from popular nodes to long-tail nodes. We also propose a novel hashing trick to reduce the redundant space consumption so that CP-LSH is capable of tackling industry-scale data. Extensive offline experiments have been conducted on three large-scale public datasets. We also deploy CP-LSH to real-world recommendation systems in one of the largest e-commerce platforms for online experiments. Experimental results demonstrate that CP-LSH outperforms competitive baseline methods in node classification and link prediction tasks. Besides, the results of online experiments also indicate that CP-LSH is practical and robust for real-world production systems.

Different from traditional machine learning tasks and benchmarks, real-world problems are usually accompanied by enormous output spaces, from hundred thousands of diseases in medical diagnosis, to millions of items and billions of websites in product and web search engines. Unfortunately, conventional machine learning tools and libraries are incapable of efficiently and accurately tackling large-scale output spaces. To address this issue, PECOS (Prediction for Enormous and Correlated Output Spaces) [11] is a state-of-the-art and open-sourced machine learning library1, which not only provides high-level and user-friendly interfaces of both linear and deep learning models, but also supplies considerable flexibility for solving diverse machine learning problems. Specifically, PECOS eases complicated semantic indexing for organizing enormous output spaces, thereby efficiently training models and deriving predictions by magnitude orders on correlated output labels. As a powerful and useful framework, PECOS has already been adopted in various real- world large-scale products like semantic search in Amazon [1], as well as achieved state-of-the-art on public extreme multi-label classification (XMC) benchmarks [2, 11, 12 ] and various downstream applications [3, 7, 9].

In the era of big data, eXtreme Multi-label Classification (XMC) has already become one of the most essential research tasks to deal with enormous label spaces in machine learning applications. Instead of assessing every individual label, most XMC methods rely on label trees or filters to derive short ranked label lists as prediction, thereby reducing computational overhead. Specifically, existing studies obtain ranked label lists with a fixed length for prediction and evaluation. However, these predictions are unreasonable since data points have varied numbers of relevant labels. The greatly small and large list lengths in evaluation, such as [email protected] and [email protected], can also lead to the ignorance of other relevant labels or the tolerance of many irrelevant labels. In this paper, we aim to provide reasonable prediction for extreme multi-label classification with dynamic numbers of predicted labels. In particular, we propose a novel framework, Model-Agnostic List Truncation with Ordinal Regression (MALTOR), to leverage the ranking properties and truncate long ranked label lists for better accuracy. Extensive experiments conducted on six large-scale real-world benchmark datasets demonstrate that MALTOR significantly outperforms statistical baseline methods and conventional ranked list truncation methods in ad-hoc retrieval with both linear and deep XMC models. The results of an ablation study also shows the effectiveness of each individual component in our proposed MALTOR.

COVID-19 has caused lasting damage to almost every domain in public health, society, and economy. To monitor the pandemic trend, existing studies rely on the aggregation of traditional statistical models and epidemic spread theory. In other words, historical statistics of COVID-19, as well as the population mobility data, become the essential knowledge for monitoring the pandemic trend. However, these solutions can barely provide precise prediction and satisfactory explanations on the long-term disease surveillance while the ubiquitous social media resources can be the key enabler for solving this problem. For example, serious discussions may occur on social media before and after some breaking events take place. To take advantage of the social media data, we propose a novel framework, Social Media enhAnced pandemic suRveillance Technique (SMART), which is composed of two modules: (i) information extraction module to construct heterogeneous knowledge graphs based on the extracted events and relationships among them; (ii) time series prediction module to provide both short-term and long-term forecasts of the confirmed cases and fatality at the state-level in the United States and to discover risk factors for COVID-19 interventions. Extensive experiments show that our method largely outperforms the state-of-the-art baselines by 7.3% and 7.4% in confirmed case/fatality prediction, respectively.

Motives or goals are recognized in psychology literature as the most fundamental drive that explains and predicts why people do what they do, including when they browse the web. Although providing enormous value, these higher-ordered goals are often unobserved, and little is known about how to leverage such goals to assist people’s browsing activities. This paper proposes to take a new approach to address this problem, which is fulfilled through a novel neural framework, Goal-directed Web Browsing (GoWeB). We adopt a psychologically-sound taxonomy of higher-ordered goals and learn to build their representations in a structure-preserving manner. Then we incorporate the resulting representations for enhancing the experiences of common activities people perform on the web. Experiments on large-scale data from Microsoft Edge web browser show that GoWeB significantly outperforms competitive baselines for in-session web page recommendation, re-visitation classification, and goal-based web page grouping. A follow-up analysis further characterizes how the variety of human motives can affect the difference observed in human behavioral patterns.

The transparency issue of sponsorship disclosure in advertising posts has become a significant problem in influencer marketing. Although influencers are urged to comply with the regulations governing sponsorship disclosure, a considerable number of influencers fail to disclose sponsorship properly in paid advertisements. In this paper, we propose a learning-to-rank based model, Sponsored Post Detector (SPoD), to detect undisclosed sponsorship of social media posts by learning various aspects of the posts such as text, image, and the social relationship among influencers and brands. More precisely, we exploit image objects and contextualized information to obtain the representations of the posts and also utilize Graph Convolutional Networks (GCNs) on a network which consists of influencers, brands, and posts with embed social media attributes. We further optimize the model by conducting manifold regularization based on temporal information and mentioned brands in posts. The extensive studies and experiments are conducted on sampled real-world Instagram datasets containing 1,601,074 posts, which mention 26,910 brands, published over 6 years by 38,113 influencers. Our experimental results demonstrate that SPoD significantly outperforms the existing baseline methods in discovering sponsored posts on social media.

MOTIVATION: Circular RNA (circRNA) is a novel class of long non-coding RNAs that have been broadly discovered in the eukaryotic transcriptome. The circular structure arises from a non-canonical splicing process, where the donor site backspliced to an upstream acceptor site. These circRNA sequences are conserved across species. More importantly, rising evidence suggests their vital roles in gene regulation and association with diseases. As the fundamental effort toward elucidating their functions and mechanisms, several computational methods have been proposed to predict the circular structure from the primary sequence. Recently, advanced computational methods leverage deep learning to capture the relevant patterns from RNA sequences and model their interactions to facilitate the prediction. However, these methods fail to fully explore positional information of splice junctions and their deep interaction. RESULTS: We present a robust end-to-end framework, Junction Encoder with Deep Interaction (JEDI), for circRNA prediction using only nucleotide sequences. JEDI first leverages the attention mechanism to encode each junction site based on deep bidirectional recurrent neural networks and then presents the novel cross-attention layer to model deep interaction among these sites for backsplicing. Finally, JEDI can not only predict circRNAs but also interpret relationships among splice sites to discover backsplicing hotspots within a gene region. Experiments demonstrate JEDI significantly outperforms state-of-the-art approaches in circRNA prediction on both isoform level and gene level. Moreover, JEDI also shows promising results on zero-shot backsplicing discovery, where none of the existing approaches can achieve. AVAILABILITY AND IMPLEMENTATION: The implementation of our framework is available at https://github.com/hallogameboy/JEDI. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Humor plays an important role in human languages and it is essential to model humor when building intelligence systems. Among different forms of humor, puns perform wordplay for humorous effects by employing words with double entendre and high phonetic similarity. However, identifying and modeling puns are challenging as puns usually involved implicit semantic or phonological tricks. In this paper, we propose Pronunciation-attentive Contextualized Pun Recognition (PCPR) to perceive human humor, detect if a sentence contains puns and locate them in the sentence. PCPR derives contextualized representation for each word in a sentence by capturing the association between the surrounding context and its corresponding phonetic symbols. Extensive experiments are conducted on two benchmark datasets. Results demonstrate that the proposed approach significantly outperforms the state-of-the-art methods in pun detection and location tasks. In-depth analyses verify the effectiveness and robustness of PCPR.

Information networks, such as social and citation networks, are ubiquitous in the real world so that network analysis plays an important role in data mining and knowledge discovery. To alleviate the sparsity problem of network analysis, it is common to capture the network semantics by projecting nodes onto a vector space as network embeddings. Moreover, random walks are usually exploited to efficiently learn node embeddings and preserve network proximity. In addition to proximity structure, heterogeneous networks have more knowledge about the types of nodes. However, to profit from heterogeneous knowledge, most of the existing approaches guide the random walks through predefined meta-paths or specific strategies, which can distort the understanding of network structures. Furthermore, traditional random walk-based approaches much favor the nodes with higher degrees while other nodes are equivalently important for the downstream applications. In this paper, we propose Meta-context Aware Random Walks (MARU) to overcome these challenges, thereby learning richer and more unbiased representations for heterogeneous networks. To reduce the bias in classical random walks, the algorithm of bidirectional extended random walks is introduced to improve the fairness of representation learning. Based on the enhanced random walks, the meta-context aware skip-gram model is then presented to learn robust network embeddings with dynamic meta-contexts. Therefore, MARU can not only fairly understand the overall network structures but also leverage the sophisticated heterogeneous knowledge in the networks. Extensive experiments have been conducted on three real-world large-scale publicly available datasets. The experimental results demonstrate that MARU significantly outperforms state-of-the-art heterogeneous network embedding methods across three general machine learning tasks, including multi-label node classification, node clustering, and link prediction.

Speaker identification answers the fundamental question "Who is speaking" The identification technology enables various downstream applications to provide a personalized experience. Both the prevalent i-vector based solutions and the state-of-the-art deep learning solutions usually treat all users equally, with no distinctions between new users and existing users, during the training process. We notice that a good many new users start with limited labeled training data, which often results in inferior predicting performance of recognizing users' voices. To alleviate the disadvantage caused by training data deficiency, we propose a Mixture Density Network- based Meta-Learning method (MDNML) for speaker identification. MDNML emphasizes the expeditious process of learning to recognize new users where each has only a few seconds of labeled data. We conduct experiments on the LibriSpeech dataset and compare MDNML with four state-of-the-art baseline methods. The results conclude that MDNML achieves higher accuracy in recognizing new users with limited labeled utterances than all baseline methods. Our proposed solution significantly expedites the learning by transferring the knowledge learned from the existing user base through gradient-based meta-learning. We consider our work to be a steppingstone for more sophisticated meta-learning frameworks for accelerating voice recognition. Furthermore, we discuss a strategy for enhancing the accuracy by incorporating the notion of household-based acoustic profiles with MDNML.

Determining user geolocation is vital to various real-world applications on the internet, such as online marketing and event detection. To identify the geolocations of users, their behaviors on social media like published posts and social interactions can be strong evidence. However, most of the existing social media based approaches individually learn from text contexts and social networks. This separation can not only lead to sub-optimal performance but also ignore the distinct importance of two resources for different users. To address this challenge, we propose a novel end-to-end framework, Hybrid-attentive User Geolocation (HUG), to jointly model post texts and user interactions in social media. The hybrid attention mechanism is introduced to automatically determine the importance of texts and social networks for each user while social media posts and interactions are modeled by a graph attention network and a language attention network. Extensive experiments conducted on three benchmark geolocation datasets using Twitter data demonstrate that HUG significantly outperforms competitive baseline methods. The in-depth analysis also indicates the robustness and interpretability of HUG.

Automatic speaker recognition (ASR) is a stepping-stone technology towards semantic multimedia understanding and benefits versatile downstream applications. In recent years, neural network-based ASR methods have demonstrated remarkable power to achieve excellent recognition performance with sufficient training data. However, it is impractical to collect sufficient training data for every user, especially for fresh users. Therefore, a large portion of users usually has a very limited number of training instances. As a consequence, the lack of training data prevents ASR systems from accurately learning users acoustic biometrics, jeopardizes the downstream applications, and eventually impairs user experience.

Top-K recommender systems aim to generate few but satisfactory personalized recommendations for various practical applications, such as item recommendation for e-commerce and link prediction for social networks. However, the numbers of users and items can be enormous, thereby leading to myriad potential recommendations as well as the bottleneck in evaluating and ranking all possibilities. Existing Maximum Inner Product Search (MIPS) based methods treat the item ranking problem for each user independently and the relationship between users has not been explored. In this paper, we propose a novel model for clustering and navigating for top-K recommenders (CANTOR) to expedite the computation of top-K recommendations based on latent factor models. A clustering-based framework is first presented to leverage user relationships to partition users into affinity groups, each of which contains users with similar preferences. CANTOR then derives a coreset of representative vectors for each affinity group by constructing a set cover with a theoretically guaranteed difference to user latent vectors. Using these representative vectors in the coreset, approximate nearest neighbor search is then applied to obtain a small set of candidate items for each affinity group to be used when computing recommendations for each user in the affinity group. This approach can significantly reduce the computation without compromising the quality of the recommendations. Extensive experiments are conducted on six publicly available large-scale real-world datasets for item recommendation and personalized link prediction. The experimental results demonstrate that CANTOR significantly speeds up matrix factorization models with high precision. For instance, CANTOR can achieve 355.1x speedup for inferring recommendations in a million-user network with 99.5% [email protected] to the original system while the state-of-the-art method can only obtain 93.7x speedup with 99.0% [email protected]

Modern online content-sharing platforms host billions of items like music, videos, and products uploaded by various providers for users to discover items of their interests. To satisfy the information needs, the task of effective item retrieval (or item search ranking) given user search queries has become one of the most fundamental problems to online content-sharing platforms. Moreover, the same query can represent different search intents for different users, so personalization is also essential for providing more satisfactory search results. Different from other similar research tasks, such as ad-hoc retrieval and product retrieval with copious words and reviews, items in content-sharing platforms usually lack sufficient descriptive information and related meta-data as features. In this paper, we propose the end-to-end deep attentive model (EDAM) to deal with personalized item retrieval for online content-sharing platforms using only discrete personal item history and queries. Each discrete item in the personal item history of a user and its content provider are first mapped to embedding vectors as continuous representations. A query-aware attention mechanism is then applied to identify the relevant contexts in the user history and construct the overall personal representation for a given query. Finally, an extreme multi-class softmax classifier aggregates the representations of both query and personal item history to provide personalized search results. We conduct extensive experiments on a large-scale real-world dataset with hundreds of million users from a large video media platform at Google. The experimental results demonstrate that our proposed approach significantly outperforms several competitive baseline methods. It is also worth mentioning that this work utilizes a massive dataset from a real-world commercial content-sharing platform for personalized item retrieval to provide more insightful analysis from the industrial aspects.

Influencer marketing has become a key marketing method for brands in recent years. Hence, brands have been increasingly utilizing influencers’ social networks to reach niche markets, and researchers have been studying various aspects of influencer marketing. However, brands have often suffered from searching and hiring the right influencers with specific interests/topics for their marketing due to a lack of available influencer data and/or limited capacity of marketing agencies. This paper proposes a multimodal deep learning model that uses text and image information from social media posts (i) to classify influencers into specific interests/topics (e.g., fashion, beauty) and (ii) to classify their posts into certain categories. We use the attention mechanism to select the posts that are more relevant to the topics of influencers, thereby generating useful influencer representations. We conduct experiments on the dataset crawled from Instagram, which is the most popular social media for influencer marketing. The experimental results show that our proposed model significantly outperforms existing user profiling methods by achieving 98% and 96% accuracy in classifying influencers and their posts, respectively. We release our influencer dataset of 33,935 influencers labeled with specific topics based on 10,180,500 posts to facilitate future research.

Question Routing (QR) on Community-based Question Answering (CQA) websites aims at recommending answerers that have high probabilities of providing the “accepted answers” to new questions. The existing question routing algorithms simply predict the ranking of users based on query content. As a consequence, the question raiser information is ignored. On the other hand, they lack learnable scoring functions to explicitly compute ranking scores.To tackle these challenges, we propose NeRank that (1) jointly learns representations of question content, question raiser, and question answerers by a heterogeneous information network embedding algorithm and a long short-term memory (LSTM) model. The embeddings of the three types of entities are unified in the same latent space, and (2) conducts question routing for personalized queries, i.e., queries with two entities (question content, question raiser), by a convolutional scoring function taking the learned embeddings of all three types of entities as input. Using the scores, NeRank routes new questions to high-ranking answerers that are skillfulness in the question domain and have similar backgrounds to the question raiser.Experimental results show that NeRank significantly outperforms competitive baseline question routing models that ignore the raiser information in three ranking metrics. In addition, NeRank is convergeable in several thousand iterations and insensitive to parameter changes, which prove its effectiveness, scalability, and robustness.

Accompanied by modern industrial developments, air pollution has already become a major concern for human health.Hence, air quality measures, such as the concentration of PM 2.5 , have attracted increasing attention.Even some studies apply historical measurements into air quality forecast, the changes of air quality conditions are still hard to monitor.In this paper, we propose to exploit social media and natural language processing techniques to enhance air quality prediction.Social media users are treated as social sensors with their findings and locations.After filtering noisy tweets using word selection and topic modeling, a deep learning model based on convolutional neural networks and overtweet-pooling is proposed to enhance air quality prediction.We conduct experiments on 7month real-world Twitter datasets in the five most heavily polluted states in the USA.The results show that our approach significantly improves air quality prediction over the baseline that does not use social media by 6.9% to 17.7% in macro-F1 scores.

Stress is a common problem in modern life that can bring both psychological and physical disorder. Wearable sensors are commonly used to study the relationship between physical records and mental status. Although sensor data generated by wearable devices provides an opportunity to identify stress in people for predictive medicine, in practice, the data are typically complicated and vague and also often fragmented. In this paper, we propose DataCompletion with Diurnal Regularizers (DCDR) and TemporallyHierarchical Attention Network (THAN) to address the fragmented data issue and predict human stress level with recovered sensor data. We model fragmentation as a sparsity issue. The nuclear norm minimization method based on the low-rank assumption is first applied to derive unobserved sensor data with diurnal patterns of human behaviors. A hierarchical recurrent neural network with the attention mechanism then models temporally structural information in the reconstructed sensor data, thereby inferring the predicted stress level. Data for this study were from 75 undergraduate students (taken from a sample of a larger study) who provided sensor data from smart wristbands. They also completed weekly stress surveys as ground-truth labels about their stress levels. This survey lasted 12 weeks and the sensor records are also in this period. The experimental results demonstrate that our approach significantly outperforms conventional methods in both data completion and stress level prediction. Moreover, an in-depth analysis further shows the effectiveness and robustness of our approach.

Yichao Zhou, Jyun-Yu Jiang, Kai-Wei Chang, Wei Wang. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Identifying and recommending potential new customers for local businesses are crucial to the survival and success of local businesses. A key component to identifying the right customers is to understand the decision-making process of choosing a business over the others. However, modeling this process is an extremely challenging task as a decision is influenced by multiple factors. These factors include but are not limited to an individual's taste or preference, the location accessibility of a business, and the reputation of a business from social media. Most of the recommender systems lack the power to integrate multiple factors together and are hardly extensible to accommodate new incoming factors. In this paper, we introduce a unified framework, CORALS, which considers the personal preferences of different customers, the geographical influence, and the reputation of local businesses in the customer recommendation task. To evaluate the proposed model, we conduct a series of experiments to extensively compare with 12 state-of-the-art methods using two real-world datasets. The results demonstrate that CORALS outperforms all these baselines by a significant margin in most scenarios. In addition to identifying potential new customers, we also break down the analysis for different types of businesses to evaluate the impact of various factors that may affect customers' decisions. This information, in return, provides a great resource for local businesses to adjust their advertising strategies and business services to attract more prospective customers.

Semantic text matching is one of the most important research problems in many domains, including, but not limited to, information retrieval, question answering, and recommendation. Among the different types of semantic text matching, long-document-to-long-document text matching has many applications, but has rarely been studied. Most existing approaches for semantic text matching have limited success in this setting, due to their inability to capture and distill the main ideas and topics from long-form text.

Conventional approaches to relevance ranking typically optimize ranking models by each query separately. The traditional cluster hypothesis also does not consider the dependency between related queries. The goal of this paper is to leverage similar search intents to perform ranking consistency so that the search performance can be improved accordingly. Different from the previous supervised approach, which learns relevance by click‐through data, we propose a novel cocluster hypothesis to bridge the gap between relevance ranking and ranking consistency. A nearest‐neighbors test is also designed to measure the extent to which the cocluster hypothesis holds. Based on the hypothesis, we further propose a two‐stage unsupervised approach, in which two ranking heuristics and a cost function are developed to optimize the combination of consistency and uniqueness (or inconsistency). Extensive experiments have been conducted on a real and large‐scale search engine log. The experimental results not only verify the applicability of the proposed cocluster hypothesis but also show that our approach is effective in boosting the retrieval performance of the commercial search engine and reaches a comparable performance to the supervised approach.

Search engine users always endeavor to reformulate queries during search sessions for articulating their information needs because it is not always easy to articulate the search intents. To further ameliorate the reformulation process, search engines may provide some query suggestions based on previous queries. In this paper, we propose Reformulation Inference Network (RIN) to learn how users reformulate queries, thereby benefiting context-aware query suggestion. Instead of categorizing reformulations into predefined patterns, we represent queries and reformulations in a homomorphic hidden space through heterogeneous network embedding. To capture the structure of the session context, a recurrent neural network (RNN) with the attention mechanism is employed to encode the search session by reading the homomorphic query and reformulation embeddings. It enables the model to explicitly captures the former reformulation for each query in the search session and directly learn user reformulation behaviors, from which query suggestion may benefit as shown in previous studies. To generate query suggestions, a binary classifier and an RNN-based decoder are introduced as the query discriminator and the query generator. Inspired by the intuition that model accurately predicting the next reformulation can also correctly infer the next intended query, a reformulation inferencer is then designed for inferring the next reformulation in the latent space of homomorphic embeddings. Therefore, both question suggestion and reformulation prediction can be simultaneously optimized by multi-task learning. Extensive experiments are conducted on publicly available AOL search engine logs. The experimental results demonstrate that RIN outperforms competitive baselines across various situations for both discriminative and generative tasks of context-aware query suggestion.

Jyun-Yu Jiang, Francine Chen, Yan-Ying Chen, Wei Wang. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

Online streaming services are prevalent. Major service providers, such as Netflix (for movies) and Spotify (for music), usually have a large customer base. More often than not, users may share an account. This has attracted increasing attention recently, as account sharing not only compromises the service provider's financial interests but also impairs the performance of recommendation systems and consequently the quality of service provided to the users. To address this issue, this paper focuses on the problem of user identification in shared accounts. Our goal is three-fold: (1) Given an account, along with its historical session logs, we identify a set of users who share such account; (2) Given a new session issued by an account, we find the corresponding user among the identified users of such account; (3) We aim to boost the performance of item recommendation by user identification. While the mapping between users and accounts is unknown, we propose an unsupervised learning-based framework, Session-based Heterogeneous graph Embedding for User Identification (SHE-UI), to differentiate and model the preferences of users in an account, and to group sessions by these users. In SHE-UI, a heterogeneous graph is constructed to represent items such as songs and their available metadata such as artists, genres, and albums. An item-based session embedding technique is proposed using a normalized random walk in the heterogeneous graph. Our experiments conducted on two large-scale music streaming datasets, Last.fm and KKBOX, show that SHE-UI not only accurately identifies users, but also significantly improves the performance of item recommendation over the state-of-the-art methods.

Social coding and open source repositories have become more and more popular. Software developers have various alternatives to contribute themselves to the communities and collaborate with others. However, nowadays there is no effective recommender suggesting developers appropriate repositories in both the academia and the industry. Although existing one-class collaborative filtering (OCCF) approaches can be applied to this problem, they do not consider particular constraints of social coding such as the programming languages, which, to some extent, associate the repositories with the developers. The aim of this paper is to investigate the feasibility of leveraging user programming language preference to improve the performance of OCCF-based repository recommendation. Based on matrix factorization, we propose language-regularized matrix factorization (LRMF), which is regularized by the relationships between user programming language preferences. Extensive experiments have been conducted on the real-world dataset of GitHub. The results demonstrate that our framework significantly outperforms five competitive baselines.

Nowadays, geosensor data, such as air quality and traffic flow, have become more and more essential in people's daily life. However, installing geosensors or hiring volunteers at every location and every time is so expensive. Some organizations may have only few facilities or limited budget to sense these data. Moreover, people usually tend to know the forecast instead of ongoing observations, but the number of sensors (or volunteers) will be a hurdle to make precise prediction. In this paper, we propose a novel concept to forecast geosensor data with participatory sensing. Given a limited number of sensors or volunteers, participatory sensing assumes each of them can observe and collect data at different locations and at different time. By aggregating these sparse data observations in the past time, we propose a neural network based approach to forecast the future geosensor data in any location of an urban area. The extensive experiments have been conducted with large-scale datasets of the air quality in three cities and the traffic of bike sharing systems in two cities. Experimental results show that our predictive model can precisely forecast the air quality and the bike rentle traffic as geosensor data.

In this paper, we propose a new idea called ranking consistency in web search. Relevance ranking is one of the biggest problems in creating an effective web search system. Given some queries with similar search intents, conventional approaches typically only optimize ranking models by each query separately. Hence, there are inconsistent rankings in modern search engines. It is expected that the search results of different queries with similar search intents should preserve ranking consistency. The aim of this paper is to learn consistent rankings in search results for improving the relevance ranking in web search. We then propose a re-ranking model aiming to simultaneously improve relevance ranking and ranking consistency by leveraging knowledge bases and search logs. To the best of our knowledge, our work offers the first solution to improving relevance rankings with ranking consistency. Extensive experiments have been conducted using the Freebase knowledge base and the large-scale query-log of a commercial search engine. The experimental results show that our approach significantly improves relevance ranking and ranking consistency. Two user surveys on Amazon Mechanical Turk also show that users are sensitive and prefer the consistent ranking results generated by our model.

Nowadays, World Wide Web connects people to each other in many ways ubiquitously. Followed along with the convenience and usability, millions of malware infect various devices of numerous users through the web every day. In contrast, traditional anti-malware systems detect such malware by scanning file systems and provide secure environments for users. However, some malware might not be detected by traditional scanning-based detection systems due to hackers' obfuscation techniques. Also, scanning-based approaches cannot caution users for uninfected malware with high risks. In this paper, we aim to build a personalized malware warning system. Different from traditional scanning-based approaches, we focus on discovering the potential malware which has not been detected for each user. If users and the system know the potentially infected malware in advance, they can be alert against the corresponding risks. We propose a novel approach to learn the implicit feedback from detection logs and give a personalized risk ranking of malware for each user. Finally, the experiments on real-world detection datasets demonstrate the proposed algorithm outperforms traditional popularity-based algorithms.

It is crucial for query auto-completion to accurately predict what a user is typing. Given a query prefix and its context (e.g., previous queries), conventional context-aware approaches often produce relevant queries to the context. The purpose of this paper is to investigate the feasibility of exploiting the context to learn user reformulation behavior for boosting prediction performance. We first conduct an in-depth analysis of how the users reformulate their queries. Based on the analysis, we propose a supervised approach to query auto-completion, where three kinds of reformulation-related features are considered, including term-level, query-level and session-level features. These features carefully capture how the users change preceding queries along the query sessions. Extensive experiments have been conducted on the large-scale query log of a commercial search engine. The experimental results demonstrate a significant improvement over 4 competitive baselines.

Microblog users often post what they observe in the surroundings, making it possible to use such microblog data to perform event detection. In this paper, we propose an issue of event duration detection and choose rain as our target event. Our goal is to construct an online virtual weather station, which reports local weather condition such as rain with the microblog data. Different from previous work focusing on earthquake, rain is a relatively minor event and may continue for a period of time. The virtual station, therefore, needs to detect not only when it starts but also when it finishes. The system trains a classifier to extract the event of interest. A wavelet-based method and aging-based method are proposed to detect the beginning of the rain event and its duration, respectively. Our experiments are conducted on real data, collected from Twitter and online weather stations. The results of experiments show the feasibility of virtual weather system. Our user behavior analysis also explains why the system works.