With the rapid development of e-commerce, auto-bidding has become a key asset in optimizing advertising performance under diverse advertiser environments. The current approaches focus on reinforcement learning (RL) and generative models. These efforts imitate offline historical behaviors by utilizing a complex structure with expensive hyperparameter tuning. The suboptimal trajectories further exacerbate the difficulty of policy learning. To address these challenges, we proposes QGA, a novel Q-value regularized Generative Auto-bidding method. In QGA, we propose to plug a Q-value regularization with double Q-learning strategy into the Decision Transformer backbone. This design enables joint optimization of policy imitation and action-value maximization, allowing the learned bidding policy to both leverage experience from the dataset and alleviate the adverse impact of the suboptimal trajectories. Furthermore, to safely explore the policy space beyond the data distribution, we propose a Q-value guided dual-exploration mechanism, in which the DT model is conditioned on multiple return-to-go targets and locally perturbed actions. This entire exploration process is dynamically guided by the aforementioned Q-value module, which provides principled evaluation for each candidate action. Experiments on public benchmarks and simulation environments demonstrate that QGA consistently achieves superior or highly competitive results compared to existing alternatives. Notably, in large-scale real-world A/B testing, QGA achieves a 3.27% increase in Ad GMV and a 2.49% improvement in Ad ROI.

Automated bidding is central to modern digital advertising. Early rule-based methods lacked adaptability, while subsequent Reinforcement Learning approaches modeled bidding as a Markov Decision Process but struggled with long-term dependencies. Recent generative models show promise, yet they lack explicit mechanisms to balance exploration and safety, relying solely on action perturbations or trajectory guidance without a safety fallback. This results in inefficient exploration and elevated financial risk for advertising platforms. To address this gap, we propose GUIDE (Generative Auto-Bidding with Unified Modeling and Exploration), a framework that synergistically integrates directed exploration with a safe fallback mechanism. GUIDE employs a Decision Transformer (DT) to jointly model historical bidding actions and environmental state transitions. A Q-value module guides the DT's exploration via regularization constraints, while an Inverse Dynamics Module (IDM) leverages DT-predicted future states to infer robust, behaviorally consistent actions as a safe policy fallback. The Q-value module then adaptively selects the final action between these two options, balancing exploration and safety. Together, these components form an integrated "explore-safeguard-select" pipeline that unifies efficiency and safety. We conduct extensive experiments on public datasets, in simulated auction environments, and through large-scale online deployment on Taobao, a leading Chinese advertising platform. Results show GUIDE consistently outperforms state-of-the-art baselines across all scenarios. In real-world deployment, GUIDE achieves notable gains: +4.10% ad GMV, +1.40% ad clicks, +1.66% ad cost, and +3.52% ad ROI, demonstrating its effectiveness and strong industrial applicability.

Scaling laws have enabled large language models(LLMs) to achieve remarkable performance and strong generalization across diverse language understanding tasks, including few-shot, in-context, and zero-shot learning. While prior studies in large-scale collaborative filtering(CF) have revealed clear relationships between model performance and scaling factors such as data size and model capacity, little attention has been given to how heterogeneous datasets can be synergistically combined for recommender systems(RS). In particular, it remains unclear whether systematically integrating diverse recommendation datasets can yield scaling behaviors analogous to those observed in LLMs, while simultaneously addressing challenges such as cold-start recommendation and cross-domain transfer. In this paper, we present RecCLIP, a multimodal framework that reformulates user--item interactions as visual representations compatible with vision--language models(VLMs). RecCLIP compresses interaction signals and employs prompt-based ranking to enable unified representation across heterogeneous data sources. Extensive experiments reveal consistent power-law scaling trends with respect to data size, and demonstrate that RecCLIP achieves superior performance in both cold-start and cross-domain transfer scenarios. Our findings underscore the importance of data-centric design in recommender systems and provide practical insights into scaling them effectively.The code for replication is available at https://github.com/jinliwei-1/RecCLIP.

Search engines, Recommender systems, and Online advertising are playing fundamental roles in modern web and mobile applications. In these information systems, the most significant component is the ranking system, which selects a list of items likely to interest a user from billions of candidate items. At its core, Deep learning to Rank (DLTR) has become indispensable for building high-performance ranking models, driving significant gains in user engagement and business growth. In this article, firstly, we outline the key problems and challenges in industrial-scale ranking systems. Secondly, we provide a comprehensive review of deep learning models deployed across multiple stages of the industrial ranking pipeline, including matching, pre-ranking, fine-grained ranking, post-ranking, and relevance-ranking. Finally, we explore novel perspectives for future research, such as leveraging Large Language Models (LLMs). The papers discussed in this survey are listed in https://github.com/guyulongcs/Awesome-Deep-Learning-Papers-for-Search-Recommendation-Advertising .

Existing recommendation methods struggle to model users’ multifaceted preferences due to the diversity and volatility of user behavior, as well as the inherent uncertainty and ambiguity of item themes in practical scenarios. Multi-interest recommendation addresses this challenge by explicitly extracting multiple interest representations from users’ historical interactions, enabling fine-grained preference modeling and more accurate recommendations. It has attracted considerable attention in recommendation research. However, current recommendation surveys have either delved into specific recommendation tasks and downstream applications or focused on approaches that model users and items as single representations with cutting-edge techniques, overlooking users’ diverse preferences and the multifaceted aspects of items. In this work, we systematically review the progress, solutions, challenges, and future directions of multi-interest recommendation by answering the following three questions: (1) Why is multi-interest modeling significantly important for recommendation? (2) What aspects are focused on by multi-interest modeling in recommendation? and (3) How can multi-interest modeling be applied, along with the technical details of the representative modules? We hope that this survey establishes a fundamental framework and delivers a preliminary overview for researchers interested in this field and committed to further exploration. The implementation of multi-interest recommendation summarized in this survey is maintained at https://github.com/WHUIR/Multi-Interest-Recommendation-A-Survey .

This is a corrigendum for the article “One Model for All: Large Language Models are Domain-Agnostic Recommendation Systems” published in ACM Trans. Inf. Syst . 43, 5, Article 118 (July 2025), 27 pages.

Recently, graph contrastive learning (GCL) has attracted considerable attention in social recommendation, owing to its ability to enhance the robustness of node embedding learning against noise and data sparsity. Despite their effectiveness, we argue that existing GCL-based methods remain limited by three key issues: (1) during graph propagation, they rely on uniform neighbor aggregation and non-adaptive embedding readout, leading to suboptimal node representations; (2) when constructing contrastive views, they typically adopt graph augmentations based on stochastic perturbations of graph-structured data, which may undermine model fidelity; (3) during model optimization, they treat all observed instances equally, forgoing the subtle difference of each positive sample at different training periods. To address these limitations, we propose an effective and efficient GCL framework (EGCL) for social recommendation. Specifically, we devise a graph adaptive propagation module to learn informative embeddings of all items and users. Furthermore, we devise an augmentation-free dual CL paradigm, which consists of intra-CL within a single domain and inter-CL between two separate domains. In addition, we develop a self-adaptive weighted supervised learning paradigm and formulate the whole training procedure as a bi-level optimization problem. Extensive experiments are performed on four benchmarks, demonstrating the effectiveness and efficiency of EGCL over recent state-of-the-art recommenders. Our implementation and datasets are available at https://github.com/wubinzzu/EGCL .

Generative retrieval presents a promising approach to information retrieval, streamlining both indexing and retrieval processes through end-to-end optimization. This method typically involves assigning a unique identifier to each document, with the retrieval goal being the generation of the correct document identifier in response to a query. Although generative retrieval has demonstrated empirical success in various tasks, designing an effective document identifier remains a challenge. Previous studies have either depended excessively on one-to-one discrete identifiers, leading to increased retrieval latency and loss of semantics in documents or have used retrieval-agnostic dense document identifiers, which can hinder performance. To this end, we propose to integrate the benefits of generative retrieval and dense retrieval using an encoder-decoder-based pre-trained language model. Particularly, the decoder, i.e., the discrete identifier, functions as a coarse retriever, effectively reducing the retrieval space in an end-to-end manner. As a complement, the encoder, i.e., the dense vector, serves as a fine-grained retriever, efficiently and precisely ranking documents in a condensed space. Accordingly, we introduce a three-stage end-to-end learning framework that optimizes identifiers and vectors. Extensive experiments reveal that the proposed method exceeds the current models in terms of effectiveness and time efficiency, across both small and larger corpus sets.

This is an erratum for the article “Learning Discrete Identifiers and Dense Vectors for Generative Retrieval” published in ACM Trans. Inf. Syst. 44, 2, Article 42 (December 2025), 24 pages.

The geographic location of social media users is crucial for understanding user behavior, optimizing advertising, and supporting location-based services such as emergency awareness and event monitoring services. However, existing Twitter user geolocation methods primarily focus on explicit social relationships between users while overlooking implicit location correlations, which affects the accuracy of user geolocation. To address this, this article proposes a Twitter user geolocation method (GeoICMF) based on implicit location correlations and multi-scale feature fusion. GeoICMF introduces a novel location association graph construction method to effectively capture implicit location correlations among users, an innovative multi-scale feature fusion model to dynamically fuse multi-scale features and generate richer user representations, and a pioneering geographic partitioning method to better adapt to user location distributions and enhance geolocation accuracy. Extensive experiments on three real-world datasets demonstrate that GeoICMF outperforms state-of-the-art baseline methods in Twitter user geolocation tasks, validating the effectiveness and superiority of the proposed method.

The geolocation of social users is crucial for understanding user behavior, optimizing advertisement placement, and enhancing public safety.However, existing methods tend to show some deficiencies when handling sparse datasets and may not fully capture the natural clustering characteristics of user locations, thereby resulting in inadequate geolocation accuracy.This paper proposes a novel social user geolocation method (MILGCN) that innovatively integrates Mean Shift Clustering, Incremental Learning, and Graph Convolutional Networks.Specifically, Mean Shift performs fine-grained clustering of user locations based on density peak characteristics, ensuring that geographically close users are grouped into the same cluster.Introducing an incremental learning mechanism into graph convolutional networks enables MILGCN to have progressive learning ability.As a result, the problem of incomplete feature extraction from sparse data is alleviated, resulting in more comprehensive user features and improved geolocation accuracy.Extensive experiments proved that the proposed method significantly outperforms the state-of-the-art baselines on the real Twitter datasets, demonstrating a substantial improvement in geolocation performance.

Session-based recommendation is gaining increasing attention due to its practical value in predicting the intents of anonymous users based on limited behaviors. Emerging efforts incorporate various side information to alleviate inherent data scarcity issues in this task, leading to impressive performance improvements. The core of side information-driven session-based recommendation is the discovery and utilization of diverse data. In this survey, we provide a comprehensive review of this task from a data-centric perspective. Specifically, this survey commences with a clear formulation of the task. This is followed by a detailed exploration of various benchmarks rich in side information that are pivotal for advancing research in this field. Afterwards, we delve into how different types of side information enhance the task, underscoring data characteristics and utility. Moreover, we discuss the usage of various side information, including data encoding, data injection, and involved techniques. A systematic review of research progress is then presented, with the taxonomy by the types of side information. Finally, we summarize the current limitations and present the future prospects of this vibrant topic.

The data sparsity problem has been a long-standing obstacle towards achieving better recommendation performance since it is miserable to estimate the user’s interests from limited historical behaviors. The pre-training paradigm, i.e., learning universal knowledge across a wide spectrum of domains, has increasingly become a new de-facto practice in many fields, especially for adaption to new domains. The merit of this superior generalizability renders it a natural choice to tackle the data sparsity problem for various recommendation scenarios. Hence, several efforts mainly follow masked language modeling or simple data augmentation via contrastive learning to build a pre-trained recommendation model. Our recent work (namely Miracle ) suggests that the common treatment utilizing the masked language modeling is not sufficient for pre-training a recommender system, since a user’s intent could be more complex than predicting the next word or item. The encouraging results demonstrate that the multi-interest modeling could significantly push the frontier of recommender system pre-training. Nevertheless, how to accommodate the temporal dynamics of the user interests seems to be underexplored under both single vector representation and multi-interest schemes. In this article, we aim to incorporate sophisticated temporal information modeling with the current advance in this line. More specifically, we extend Miracle by further considering relative position information and two kinds of relative time interval information jointly when performing multi-interest learning. Then, a sequential process for interest refinement is proposed to learn the subtle nuances of how interests change and shift along the timeline, leading to a more precise representation of user interests. Our extensive experiments on multiple real-world datasets validate the effectiveness of the proposed solution, demonstrating a significant improvement over current state-of-the-art models on these benchmarks. The code is available at https://github.com/WHUIR/Horae .

In legal case retrieval, existing work has shown that human-mediated conversational search can improve users’ search experience. One of the key problems for a practical conversational search system is how to ask high-quality clarifying questions to initiate conversations with users and understand their search intents. Previous works demonstrated that human-annotated external domain knowledge (such as event schemas) can improve the legal utility of clarifying questions generated by large language models. However, these methods are restricted to specific law systems or languages and cannot be generalized to others. To this end, we propose to generate context and domain-specific questions with LLMs without external annotations or knowledge by extracting information from top-retrieved documents given the current conversation context. Specifically, we construct a conversational legal case retrieval system CARQ that iteratively selects neighbor candidate case documents from the retrieved list at each conversation step to ask clarifying questions. We pretrain CARQ to capture the differences between legal cases and employ the reward augmented maximum likelihood to optimize the system directly for retrieval metrics. Extensive automated and human evaluations on three widely adopted legal case retrieval datasets demonstrate the superior effectiveness of our approach as compared with the state-of-the-art baselines.

Legal case retrieval is an important task in information retrieval that aims to retrieve relevant cases for given query cases. Conversational search paradigms have been shown to improve the search experience in legal case retrieval. However, there are two challenges in applying conversational search to legal scenarios. Firstly, legal search conversations often focus on different parts of legal case documents, but existing models struggle to capture the complex structural information and extract accurate relevance signals. Secondly, collecting large-scale conversational search datasets is costly, making it difficult to build reliable conversational legal case retrieval models. To address these challenges, we propose a Structure-Aware Matching Model (SAMM) for conversational legal case retrieval. SAMM extracts matching signals between conversational utterances and segments of the legal cases to incorporate structural information. We decouple the conversational search task into three subtasks and design pre-training tasks to overcome the lack of training data. Additionally, we create ConvLegal, the largest conversational legal case retrieval dataset to the best of our knowledge, for better evaluation of different methods. We train and evaluate SAMM and baselines on both a public dataset (CLCR) and ConvLegal. Experimental results demonstrate that SAMM outperforms existing models in legal case retrieval and conversational search.

Sequential recommendation systems aim to predict users’ next likely interaction based on their history. However, these systems face data sparsity and cold-start problems. Utilizing data from other domains, known as multi-domain methods, is useful for alleviating these problems. However, traditional multi-domain methods rely on meaningless ID-based item representation, which makes it difficult to align items with similar meanings from different domains, yielding sup-optimal knowledge transfer. This article introduces LLM-Rec , a framework that utilizes pre-trained Large Language Models (LLMs) for domain-agnostic recommendation. Specifically, we mix user’s behaviors from multiple domains and concatenate item titles into a sentence, then use LLMs for generating user and item representations. By mixing behaviors across different domains, we can exploit the knowledge encoded in LLMs to bridge the semantic across over multi-domain behaviors, thus obtaining semantically rich representations and improving performance in all domains. Furthermore, we explore the underlying reasons why LLMs are effective and investigate whether LLMs can understand the semantic correlations as the recommendation model, and if advanced techniques like scaling laws in NLP also work in recommendations. We conduct extensive experiments with LLMs ranging from 40 M to 6.7 B to answer the above questions and to verify the effectiveness of LLM-Rec in multi-domain recommendation. The source code is available at https://github.com/WHUIR/LLMRec .

Recently, the paradigm of pre-training and fine-tuning has achieved impressive performance owing to their ability to transfer general knowledge from pre-trained domain to target domain. Meanwhile, graph neural networks (GNNs) have gained prominence in recommender systems. However, there is a lack of unified pre-training and fine-tuning paradigms in graph-based recommendation systems. Applying pre-training and fine-tuning in graph-based recommendation is challenging due to the unique characteristics of recommendation data, including the non-uniform representation, negative transfer effects, and skewed data distributions. To overcome these challenges, we introduce pre-training and prompting recommendation ( ProRec ) , a novel model that synergizes uniform graph pre-training with prompt-tuning for recommendation systems. Specifically, to address the challenge of inconsistent features across different recommendation datasets, ProRec constructs unified input features at the subgraph level and uses a graph auto-encoder for pre-training, laying the foundation for uniform knowledge transfer from the pre-trained domain to the downstream domain. Additionally, ProRec employs prompt-tuning during the fine-tuning phase, which, in a parameter-efficient manner, enhances the generalization of pre-trained knowledge to downstream tasks thereby reducing negative transfer effects. Furthermore, a cross-layer contrastive learning strategy is adopted to eliminate uneven data distribution, promoting more evenly distributed and informative representations. Finally, extensive benchmark comparisons have demonstrated that ProRec outperforms the latest state-of-the-art methods. The source code necessary for replication is available at https://github.com/Code2Q/ProRec .

Recent works demonstrate the effectiveness of multi-modal information for sequential recommendation. However, the computational cost and representation degeneration fail to be focused specifically and addressed adequately in multi-modality recommendation. To this end, we first identify and formalize three properties i.e., diversity, compactness, and consistency from the geometric space and spectrum perspective. Building upon this foundation, we devise tailored loss functions to regularize the above three properties for representation optimization. Theoretical underpinnings and experimental results demonstrate the efficacy of an enhanced item representation in ameliorating degeneration. Furthermore, we propose an efficient and expandable image-centered method, named E2 ImgRec, to mitigate the immense cost of computation. Concretely, we substitute the linear projection operations in the self-attention module and feed-forward network layer with two learnable rescaling vectors or efficient recommendation, then leverage cross-attention for multi-modality information fusion. Extensive experiments on three public datasets illustrate our method outperforms representative ID-based solutions and multi-modal based state-of-the-arts with only up to 39.9% in memory usage and 4.3× acceleration in training time. The code for replication is available at https://github.com/WHUIR/E2ImgRec.

Recently, the item textual information has been exploited with pre-trained language models (PLMs) to enrich the representations of tail items. The underlying idea is to align the hot items and tail items in terms of the external semantic knowledge covered by the PLM. However, it is non-trivial to eliminate the popularity bias by exploiting the textual semantics. One major obstacle is that the model supervision still counts on the sparse yet binary user behaviors. In the preliminary investigation, we discover that text-based recommendations also suffer from the popularity bias.

Online learning to rank (OLTR), which directly optimizes the ranker with interactive user feedback, has gained considerable attention in both academia and industry. However, most current approaches suffer from the inefficiency of heuristic exploration strategies, which can seriously hurt users' experience. Furthermore, the existing OLTR solutions fail to learn from the cost-effective logged data, blocking their usage in the real industrial system. To handle the above issues, we in this paper introduce a new OLTR framework LT2R, namely Learning To online Learning to Rank. LT2R aims to study an efficient parameterized exploration strategy, by which a ranker could converge to the optimal ranking with as few exploration steps as possible. Specifically, we formulate the OLTR task as a typical Markov Decision Process and introduce an online reinforcement learning algorithm with a multi-round cumulative reward to guarantee fast convergence. Moreover, we contribute an offline learning algorithm for LT2R to exploit the knowledge from the historical searching logs, which can provide a fair warm-up model for its industrial deployment. Extensive experiments on both benchmark datasets and Baidu search engine have demonstrated its superiority over state-of-the-art methods.

Click-through rate (CTR) prediction plays a crucial role in recommendation systems, with significant impact on user experience and platform revenue generation. Despite the various public CTR datasets available due to increasing interest from both academia and industry, these datasets have limitations. They cover a limited range of scenarios and predominantly focus on ID-based features, neglecting the vital role of multi-modal features for effective multi-scenario CTR prediction. Moreover, their scale is modest compared to real-world industrial datasets, hindering robust and comprehensive evaluation of complex models. To address these challenges, we introduce a large-scale Multi-Scenario Multi-Modal CTR dataset named AntM2 C, built from real industrial data from Alipay. This dataset offers an impressive breadth and depth of information, covering CTR data from four diverse business scenarios, including advertisements, consumer coupons, mini-programs, and videos. Unlike existing datasets, AntM2 C provides not only ID-based features but also five textual features and one image feature for both users and items, supporting more delicate multi-modal CTR prediction. AntM2 C is also substantially larger than existing datasets, comprising 100 million CTR data. This scale allows for robust and comprehensive evaluation and comparison of CTR prediction models. We employ AntM2 C to construct several typical CTR tasks, including multi-scenario modeling, item and user cold-start modeling, and multi-modal modeling. Initial experiments and comparisons with baseline methods have shown that AntM2 C presents both new challenges and opportunities for CTR models, with the potential to significantly advance CTR research. The AntM2 C dataset is available at https://www.atecup.cn/OfficalDataSet.

The logs of the use of a search engine provide sufficient data to train a better ranker. However, it is well known that such implicit feedback reflects biases, and in particular a presentation bias that favors higher-ranked results. Unbiased Learning-to-Rank (ULTR) methods attempt to optimize performance by jointly modeling this bias along with the ranker so that the bias can be removed. Such methods have been shown to provide theoretical soundness, and promise superior performance and low deployment costs. However, existing ULTR methods don't recognize that query-document relevance is a confounder -- it affects both the likelihood of a result being clicked because of relevance and the likelihood of the result being ranked high by the base ranker. Moreover, the performance guarantees of existing ULTR methods assume the use of a weak ranker -- one that does a poor job of ranking documents based on relevance to a query. In practice, of course, commercial search engines use highly tuned rankers, and desire to improve upon them using the implicit judgments in search logs. This results in a significant correlation between position and relevance, which leads existing ULTR methods to overestimate click propensities in highly ranked results, reducing ULTR's effectiveness. This paper is the first to demonstrate the problem of propensity overestimation by ULTR algorithms, based on a causal analysis. We develop a new learning objective based on a backdoor adjustment. In addition, we introduce the Logging-Policy-aware Propensity (LPP) model that can jointly learn LPP and a more accurate ranker. We extensively test our approach on two public benchmark tasks and show that our proposal is effective, practical and significantly outperforms the state of the art.

Deep neural networks now have become the de-facto standard for sequential recommendation. In the existing techniques, an embedding vector is assigned for each item, encoding all the characteristics of the latter in latent space. Then, the recommendation is transferred to devising a similarity metric to recommend user's next behavior. Here, we consider each dimension of an embedding vector as a (latent) feature. Though effective, it is unknown which feature carries what semantics toward the item. Actually, in reality, this merit is highly preferable since a specific group of features could induce a particular relation among the items while the others are in vain. Unfortunately, the previous treatment overlooks the feature semantic learning at such a fine-grained level. When each item contains multiple latent aspects, which however is prevalent in real-world, the relations between items are very complex. The existing solutions are easy to fail on better recommendation performance. It is necessary to disentangle the item embeddings and extract credible features in a context-aware manner.

In the world of E-Commerce, the core task is to understand the personalized preference from various kinds of heterogeneous information, such as textual reviews, item images and historical behaviors. In current systems, these heterogeneous information are mainly exploited to generate better item or user representations. For example, in scenario of visual search, the importance of modeling query image has been widely acknowledged. But, these existing solutions focus on improving the representation quality of the query image, overlooking the personalized visual preference of the user. Note that the visual features affect the user's decision significantly, e.g., the user could be more likely to click the items with her preferred design. Hence, it is fruitful to exploit the visual preference to deliver better capacity for personalization.

In the domain of complex networks, the identification of influential nodes plays a crucial role in ensuring network stability and facilitating efficient information dissemination . Although the study of influential nodes has been applied in many fields such as suppression of rumor spreading, regulation of group behavior , and prediction of mass events evolution, current deep learning-based algorithms have limited input features and are incapable of aggregating neighbor information of nodes, thus failing to adapt to complex networks. We propose an influential node identification method in complex networks based on the Transformer. In this method, the input sequence of a node includes information about the node itself and its neighbors, enabling the model to effectively aggregate node information to identify its influence. Experiments were conducted on 9 synthetic networks and 12 real networks. Using the SIR model and a benchmark method to verify the effectiveness of our approach. The experimental results show that this method can more effectively identify influential nodes in complex networks. In particular, the method improves 27 percent compared to the second place method in network Netscience and 21 percent in network Faa.

Session-based recommendation is devoted to characterizing preferences of anonymous users based on short sessions. Existing methods mostly focus on mining limited item co-occurrence patterns exposed by item ID within sessions, while ignoring what attracts users to engage with certain items is rich multi-modal information displayed on pages. Generally, the multi-modal information can be classified into two categories: descriptive information (e.g., item images and description text) and numerical information (e.g., price). In this paper, we aim to improve session-based recommendation by modeling the above multi-modal information holistically. There are mainly three issues to reveal user intent from multi-modal information: (1) How to extract relevant semantics from heterogeneous descriptive information with different noise? (2) How to fuse these heterogeneous descriptive information to comprehensively infer user interests? (3) How to handle probabilistic influence of numerical information on user behaviors? To solve above issues, we propose a novel multi-modal session-based recommendation (MMSBR) that models both descriptive and numerical information under a unified framework. Specifically, a pseudo-modality contrastive learning is devised to enhance the representation learning of descriptive information. Afterwards, a hierarchical pivot transformer is presented to fuse heterogeneous descriptive information. Moreover, we represent numerical information with Gaussian distribution and design a Wasserstein self-attention to handle the probabilistic influence mode. Extensive experiments on three real-world datasets demonstrate the effectiveness of the proposed MMSBR. Further analysis also proves that our MMSBR can alleviate the cold-start problem in SBR effectively.

Mainstream solutions to sequential recommendation represent items with fixed vectors. These vectors have limited capability in capturing items’ latent aspects and users’ diverse preferences. As a new generative paradigm, diffusion models have achieved excellent performance in areas like computer vision and natural language processing. To our understanding, its unique merit in representation generation well fits the problem setting of sequential recommendation. In this article, we make the very first attempt to adapt the diffusion model to sequential recommendation and propose DiffuRec for item representation construction and uncertainty injection. Rather than modeling item representations as fixed vectors, we represent them as distributions in DiffuRec , which reflect a user’s multiple interests and an item’s various aspects adaptively. In the diffusion phase, DiffuRec corrupts the target item embedding into a Gaussian distribution via noise adding, which is further applied for sequential item distribution representation generation and uncertainty injection. Afterward, the item representation is fed into an approximator for target item representation reconstruction. In the reverse phase, based on a user’s historical interaction behaviors, we reverse a Gaussian noise into the target item representation, then apply a rounding operation for target item prediction. Experiments over four datasets show that DiffuRec outperforms strong baselines by a large margin. 1

Existing sequential Point of Interest (POI) recommendation methods overlook a fact that each city exhibits distinct characteristics and totally ignore the city signature. In this study, we claim that city matters in sequential POI recommendation and fully exploring city signature can highlight the characteristics of each city and facilitate cross-city complementary learning. To this end, we consider the two-city scenario and propose a Dual-Target Cross-City Sequential POI Recommendation model (DCSPR) to achieve the purpose of complementary learning across cities. On one hand, DCSPR respectively captures geographical and cultural characteristics for each city by mining intra-city regions and intra-city functions of POIs. On the other hand, DCSPR builds a transfer channel between cities based on intra-city functions, and adopts a novel transfer strategy to transfer useful cultural characteristics across cities by mining inter-city functions of POIs. Moreover, to utilize these captured characteristics for sequential POI recommendation, DCSPR involves a new region- and function-aware network for each city to learn transition patterns from multiple views. Extensive experiments conducted on two real-world datasets with four cities demonstrate the effectiveness of DCSPR .

User behavior data, such as ratings and clicks, has been widely used to build personalizing models for recommender systems. However, many unflattering factors (e.g., popularity, ranking position, users’ selection) significantly affect the performance of the learned recommendation model. Most existing work on unbiased recommendation addressed these biases from sample granularity (e.g., sample reweighting, data augmentation) or from the perspective of representation learning (e.g., bias-modeling). However, these methods are usually designed for a specific bias, lacking the universal capability to handle complex situations where multiple biases co-exist. Besides, rare work frees itself from laborious and sophisticated debiasing configurations (e.g., propensity scores, imputed values, or user behavior-generating process). Towards this research gap, in this article, we propose a universal G enerative framework for B ias D isentanglement termed as GBD , constantly generating calibration perturbations for the intermediate representations during training to keep them from being affected by the bias. Specifically, a bias-identifier that tries to retrieve the bias-related information from the representations is first introduced. Subsequently, the calibration perturbations are generated to significantly deteriorate the bias-identifier’s performance, making the bias gradually disentangled from the calibrated representations. Therefore, without relying on notorious debiasing configurations, a bias-agnostic model is obtained under the guidance of the bias identifier. We further present its universality by subsuming the representative biases and their mixture under the proposed framework. Finally, extensive experiments on the real-world, synthetic, and semi-synthetic datasets have demonstrated the superiority of the proposed approach against a wide range of recommendation debiasing methods. The code is available at https://github.com/Zhidan-Wang/GBD .

Session-based recommendation intends to predict next purchased items based on anonymous behavior sequences. Numerous economic studies have revealed that item price is a key factor influencing user purchase decisions. Unfortunately, existing methods for session-based recommendation only aim at capturing user interest preference, while ignoring user price preference. Actually, there are primarily two challenges preventing us from accessing price preference. First, the price preference is highly associated to various item features (i.e., category and brand), which asks us to mine price preference from heterogeneous information. Second, price preference and interest preference are interdependent and collectively determine user choice, necessitating that we jointly consider both price and interest preference for intent modeling. To handle above challenges, we propose a novel approach Bi-Preference Learning Heterogeneous Hypergraph Networks (BiPNet) for session-based recommendation. Specifically, the customized heterogeneous hypergraph networks with a triple-level convolution are devised to capture user price and interest preference from heterogeneous features of items. Besides, we develop a Bi-Preference Learning schema to explore mutual relations between price and interest preference and collectively learn these two preferences under the multi-task learning architecture. Extensive experiments on multiple public datasets confirm the superiority of BiPNet over competitive baselines. Additional research also supports the notion that the price is crucial for the task.

Recommender systems play a pivotal role in navigating users through vast reservoirs of information. However, data sparseness can compromise recommendation accuracy, making it challenging to improve recommendation performance. To address this issue, researchers have explored incorporating multiple data types. Yet, this approach can introduce noise that impairs the recommendations' accuracy. Therefore, it is crucial to denoise the data to enhance recommendation quality. This tutorial highlights the importance of data denoising metrics for improving the accuracy and quality of recommendations. Four groups of data denoising metrics are introduced: feature, item, pattern, and modality level. For each group, various denoising methods are presented. The tutorial emphasizes the significance of selecting the right data denoising methods to enhance recommendation quality. It provides valuable guidance for practitioners and researchers implementing reliable data denoising metrics in recommender systems. Finally, the tutorial proposes open research questions for future studies, making it a valuable resource for the research community.

Legal case retrieval is a special IR task aiming to retrieve supporting cases for a given query case. Existing works have shown that conversational search paradigm can improve users' search experience in legal case retrieval. One of the keys to a practical conversational search system is how to ask high-quality clarifying questions to initiate conversations with users and understand their search intents. Recently, Large Language Models, such as ChatGPT and GPT-4, have shown superior ability in both open-domain QA and conversations with human. Thus it is natural to believe that they could be applied to legal conversational search as well. However, our preliminary study has shown that generating clarifying questions in legal conversational search with SOTA LLMs (e.g., GPT-4) often suffers from several problems such as duplication and low-utility contents. To address these problems, we propose LeClari, which leverages legal event schema as external knowledge to instruct LLMs to generate effective clarifying questions for legal conversational search. LeClari is constructed with a prompt module and a novel legal event selection module. The former defines a prompt with legal events for clarifying question generation and the latter selects potential event types by modeling the relationships of legal event types, conversational context, and candidate cases. We also propose ranking-oriented rewards and employ the reward augmented maximum likelihood (RAML) method to optimize LeClari directly based on the final retrieval performance of the conversational legal search system. Empirical results over two widely adopted legal case retrieval datasets demonstrate the effectiveness of our approach as compared with the state-of-the-art baselines.

Named entity recognition (NER) is the task to identify text spans that mention named entities, and to classify them into predefined categories such as person, location, organization, etc. In recent years, deep learning, empowered by continuous real-valued vector representations and semantic composition through nonlinear processing, has been employed in NER systems, yielding stat-of-the-art performance. In our TKDE paper, we provide a comprehensive review on existing deep learning techniques for NER. We first introduce NER resources, including tagged NER corpora and off-the-shelf NER tools. Then, we systematically categorize existing solutions based on a taxonomy along three axes: distributed representations for input, context encoder, and tag decoder. Next, we survey the most representative methods for deep learning in new NER problem settings and applications. Finally, we present readers with the challenges faced by NER systems and outline future directions in this area.

The pre-training paradigm, i.e., learning universal knowledge across a wide spectrum of domains, has increasingly become a new de-facto practice in many fields, especially for transferring to new domains. The recent progress includes universal pre-training solutions for recommendation. However, we argue that the common treatment utilizing the masked language modeling or simple data augmentation via contrastive learning is not sufficient for pre-training a recommender system, since a user's intent could be more complex than predicting the next word or item. It is more intuitive to go a step further by devising the multi-interest driven pre-training framework for universal user understanding. Nevertheless, incorporating multi-interest modeling in recommender system pre-training is non-trivial due to the dynamic, contextual, and temporary nature of the user interests, particularly when the users are from different domains. The limited effort on this line has greatly rendered it as an open question.

Recently, Multi-Scenario Learning (MSL) is widely used in recommendation and retrieval systems in the industry because it facilitates transfer learning from different scenarios, mitigating data sparsity and reducing maintenance cost. These efforts produce different MSL paradigms by searching more optimal network structure, such as Auxiliary Network, Expert Network, and Multi-Tower Network. It is intuitive that different scenarios could hold their specific characteristics, activating the user's intents quite differently. In other words, different kinds of auxiliary features would bear varying importance under different scenarios. With more discriminative feature representations refined in a scenario-aware manner, better ranking performance could be easily obtained without expensive search for the optimal network structure. Unfortunately, this simple idea is mainly overlooked but much desired in real-world systems.

Self-supervised learning (SSL) has been recently applied to sequential recommender systems to provide high-quality user representations. However, while facilitating the learning process recommender systems, SSL is not without security threats: carefully crafted inputs can poison the pre-trained models driven by SSL, thus reducing the effectiveness of the downstream recommendation model. This work shows that poisoning attacks against the pre-training stage threaten sequential recommender systems. Without any background knowledge of the model architecture and parameters, nor any API queries, our strategy proves the feasibility of poisoning attacks on mainstream SSL-based recommender schemes as well as on commonly used datasets. By injecting only a tiny amount of fake users, we get the target item recommended to real users more than thousands of times as before, demonstrating that recommender systems have a new attack surface due to SSL. We further show our attack is challenging for recommendation platforms to detect and defend. Our work highlights the weakness of self-supervised recommender systems and shows the necessity for researchers to be aware of this security threat. Our source code is available at https://github.com/CongGroup/Poisoning-SSL-based-RS.

Candidate retrieval is a key part of the modern search engines whose goal is to find candidate items that are semantically related to the query from a large item pool. The core difference against the later ranking stage is the requirement of low latency. Hence, two-tower structure with two parallel yet independent encoder for both query and item is prevalent in many systems. In these efforts, the semantic information of a query and a candidate item is fed into the corresponding encoder and then use their representations for retrieval. With the popularity of pre-trained semantic models, the state-of-the-art for semantic retrieval tasks has achieved the significant performance gain.

Aspect-based sentiment analysis (ABSA) aims at automatically inferring the specific sentiment polarities toward certain aspects of products or services behind the social media texts or reviews, which has been a fundamental application to the real-world society. Since the early 2010s, ABSA has achieved extraordinarily high accuracy with various deep neural models. However, existing ABSA models with strong in-house performances may fail to generalize to some challenging cases where the contexts are variable, i.e., low robustness to real-world environments. In this study, we propose to enhance the ABSA robustness by systematically rethinking the bottlenecks from all possible angles, including model, data, and training. First, we strengthen the current best-robust syntax-aware models by further incorporating the rich external syntactic dependencies and the labels with aspect simultaneously with a universal-syntax graph convolutional network. In the corpus perspective, we propose to automatically induce high-quality synthetic training data with various types, allowing models to learn sufficient inductive bias for better robustness. Last, we based on the rich pseudo data perform adversarial training to enhance the resistance to the context perturbation and meanwhile employ contrastive learning to reinforce the representations of instances with contrastive sentiments. Extensive robustness evaluations are conducted. The results demonstrate that our enhanced syntax-aware model achieves better robustness performances than all the state-of-the-art baselines. By additionally incorporating our synthetic corpus, the robust testing results are pushed with around 10% accuracy, which are then further improved by installing the advanced training strategies. In-depth analyses are presented for revealing the factors influencing the ABSA robustness.

Recommender models are hard to evaluate, particularly under offline setting. In this article, we provide a comprehensive and critical analysis of the data leakage issue in recommender system offline evaluation. Data leakage is caused by not observing global timeline in evaluating recommenders e.g., train/test data split does not follow global timeline. As a result, a model learns from the user-item interactions that are not expected to be available at the prediction time. We first show the temporal dynamics of user-item interactions along global timeline, then explain why data leakage exists for collaborative filtering models. Through carefully designed experiments, we show that all models indeed recommend future items that are not available at the time point of a test instance, as the result of data leakage. The experiments are conducted with four widely used baseline models—BPR, NeuMF, SASRec, and LightGCN, on four popular offline datasets—MovieLens-25M, Yelp, Amazon-music, and Amazon-electronic, adopting leave-last-one-out data split. 1 We further show that data leakage does impact models’ recommendation accuracy. Their relative performance orders thus become unpredictable with different amount of leaked future data in training. To evaluate recommendation systems in a realistic manner in offline setting, we propose a timeline scheme, which calls for a revisit of the recommendation model design.

The Point-of-Interest (POI) transition behaviors could hold absolute sparsity and relative sparsity very differently for different cities. Hence, it is intuitive to transfer knowledge across cities to alleviate those data sparsity and imbalance problems for next POI recommendation. Recently, pre-training over a large-scale dataset has achieved great success in many relevant fields, like computer vision and natural language processing. By devising various self-supervised objectives, pre-training models can produce more robust representations for downstream tasks. However, it is not trivial to directly adopt such existing pre-training techniques for next POI recommendation, due to thelacking of common semantic objects (users or items) across different cities. Thus in this paper, we tackle such a new research problem ofpre-training across different citiesfor next POI recommendation. Specifically, to overcome the key challenge that different cities do not share any common object, we propose a novel pre-training model namedCATUS, by transferring thecategory-leveluniversal transition knowledge over different cities. Firstly, we build two self-supervised objectives inCATUS:next category predictionandnext POI prediction, to obtain the universal transition-knowledge across different cities and POIs. Then, we design acategory-transition oriented sampleron the data level and animplicit and explicit transfer strategyon the encoder level to enhance this transfer process. At the fine-tuning stage, we propose adistance oriented samplerto better align the POI representations into the local context of each city. Extensive experiments on two large datasets consisting of four cities demonstrate the superiority of our proposedCATUSover the state-of-the-art alternatives. The code and datasets are available at https://github.com/NLPWM-WHU/CATUS.

The notion personalization lies on the core of a real-world product search system, whose aim is to understand the user's search intent in a fine-grained level. The existing solutions mainly achieve this purpose through a coarse-grained semantic matching in terms of the query and item's description or the collective click correlations. Besides the issued query, the historical search behaviors of a user would cover lots of her personalized interests, which is a promising avenue to alleviate the semantic gap between users, items and queries. However, as to a specific domain, a user's search behaviors are generally sparse or even unavailable (i.e., cold-start users). How to exploit the search behaviors from the other relevant domain and enable effective fine-grained intent understanding remains largely unexplored for product search. Moreover, the semantic gap could be further aggravated since the properties of an item could evolve over time (e.g., the price adjustment for a mobile phone or the business plan update for a financial item), which is also mainly overlooked by the existing solutions.

Learning dynamic user preference has become an increasingly important component for many online platforms (e.g., video-sharing sites, e-commerce systems) to make sequential recommendations. Previous works have made many efforts to model item-item transitions over user interaction sequences, based on various architectures, e.g., recurrent neural networks and self-attention mechanism. Recently emerged graph neural networks also serve as useful backbone models to capture item dependencies in sequential recommendation scenarios. Despite their effectiveness, existing methods have far focused on item sequence representation with singular type of interactions, and thus are limited to capture dynamic heterogeneous relational structures between users and items (e.g., page view, add-to-favorite, purchase). To tackle this challenge, we design a Multi-Behavior Hypergraph-enhanced T ransformer framework (MBHT) to capture both short-term and long-term cross-type behavior dependencies. Specifically, a multi-scale Transformer is equipped with low-rank self-attention to jointly encode behavior-aware sequential patterns from fine-grained and coarse-grained levels. Additionally,we incorporate the global multi-behavior dependency into the hypergraph neural architecture to capture the hierarchical long-range item correlations in a customized manner. Experimental results demonstrate the superiority of our MBHT over various state-of- the-art recommendation solutions across different settings. Further ablation studies validate the effectiveness of our model design and benefits of the new MBHT framework. Our implementation code is released at: https://github.com/yuh-yang/MBHT-KDD22.

Fine-grained sentiment classification (FGSC) task and fine-grained controllable text generation (FGSG) task are two representative applications of sentiment analysis, two of which together can actually form an inverse task prediction, i.e., the former aims to infer the fine-grained sentiment polarities given a text piece, while the latter generates text content that describes the input fine-grained opinions. Most of the existing work solves the FGSC and the FGSG tasks in isolation, while ignoring the complementary benefits in between. This paper combines FGSC and FGSG as a joint dual learning system, encouraging them to learn the advantages from each other. Based on the dual learning framework, we further propose decoupling the feature representations in two tasks into fine-grained aspect-oriented opinion variables and content variables respectively, by performing mutual disentanglement learning upon them. We also propose to transform the difficult "data-to-text'' generation fashion widely used in FGSG into an easier text-to-text generation fashion by creating surrogate natural language text as the model inputs. Experimental results on 7 sentiment analysis benchmarks including both the document-level and sentence-level datasets show that our method significantly outperforms the current strong-performing baselines on both the FGSC and FGSG tasks. Automatic and human evaluations demonstrate that our FGSG model successfully generates fluent, diverse and rich content conditioned on fine-grained sentiments.

Recommender systems have become a fundamental service in most E-Commerce platforms, in which the matching stage aims to retrieve potentially relevant candidate items to users for further ranking. Recently, some efforts on extracting multi-interests from user's historical behaviors have demonstrated superior performance. However, the historical behaviors are not noise-free due to the possible misclicks or disturbances. Existing works mainly overlook the fact that the interests of a user are not only reflected by the historical behaviors, but also inherently regulated by the profile information. Hence, we are interested in exploiting the benefit of user profile in multi-interest learning to enhance candidate matching performance. To this end, a user-aware multi-interest learning framework (named UMI) is proposed in this paper to exploit both user profile and behavior information for candidate matching. Specifically, UMI consists of two main components: dual-attention routing and interest refinement. In the dual-attention routing, we firstly introduce a user-guided attention network to identify the important historical items with respect to the user profile. Then, the resultant importance weights are leveraged via the dual-attentive capsule network to extract the user's multi-interests. Afterwards, the extracted interests are utilized to highlight the corresponding user profile features for interest refinement, such that different user profiles can be incorporated into interest learning for diverse user preference understanding. Besides, to improve the model's discriminative capacity, we further devise a harder-negatives strategy to support model optimization. Extensive experiments show that UMI significantly outperforms state-of-the-art multi-interest modeling alternatives. Currently, UMI has been successfully deployed at Taobao App in Alibaba, serving hundreds of millions of users.

Recommender system usually faces popularity bias. From the popularity distribution shift perspective, the normal paradigm trained on exposed items (most are hot items) identifies that recommending popular items more frequently can achieve lower loss, thus injecting popularity information into item property embedding, e.g., id embedding. From the long-tail distribution shift perspective, the sparse interactions of long-tail items lead to insufficient learning of them. The resultant distribution discrepancy between hot and long-tail items would not only inherit the bias, but also amplify the bias. Existing work addresses this issue with inverse propensity scoring (IPS) or causal embeddings. However, we argue that not all popularity biases mean bad effects, i.e., some items show higher popularity due to better quality or conform to current trends, which deserve more recommendations. Blindly seeking unbiased learning may inhibit high-quality or fashionable items. To make better use of the popularity bias, we propose a co-training disentangled domain adaptation network (CD$^2$AN), which can co-train both biased and unbiased models. Specifically, for popularity distribution shift, CD$^2$AN disentangles item property representation and popularity representation from item property embedding. For long-tail distribution shift, we introduce additional unexposed items (most are long-tail items) to align the distribution of hot and long-tail item property representations. Further, from the instances perspective, we carefully design the item similarity regularization to learn comprehensive item representation, which encourages item pairs with more effective co-occurrences patterns to have more similar item property representations. Based on offline evaluations and online A/B tests, we show that CD$^2$AN outperforms the existing debiased solutions. Currently, CD$^2$AN has been successfully deployed at Mobile Taobao App and handling major online traffic.

Session-based recommendation (SBR) aims to predict a user's next clicked item based on an anonymous yet short interaction sequence. Previous SBR models, which rely only on the limited short-term transition information without utilizing extra valuable knowledge, have suffered a lot from the problem of data sparsity. This paper proposes a novel mirror graph enhanced neural model for session-based recommendation (MGS), to exploit item attribute information over item embeddings for more accurate preference estimation.

For sequential recommender, the coarse-grained yet sparse sequential signals mined from massive user-item interactions have become the bottleneck to further improve the recommendation performance. To alleviate the spareness problem, exploiting auxiliary semantic features (\eg textual descriptions, visual images and knowledge graph) to enrich contextual information then turns into a mainstream methodology. Though effective, we argue that these different heterogeneous features certainly include much noise which may overwhelm the valuable sequential signals, and therefore easily reach the phenomenon of negative collaboration (ie 1 + 1 > 2). How to design a flexible strategy to select proper auxiliary information and alleviate the negative collaboration towards a better recommendation is still an interesting and open question. Unfortunately, few works have addressed this challenge in sequential recommendation.

Sequential recommendation aims at identifying the next item that is preferred by a user based on their behavioral history. Compared to conventional sequential models that leverage attention mechanisms and RNNs, recent efforts mainly follow two directions for improvement: multi-interest learning and graph convolutional aggregation. Specifically, multi-interest methods such as ComiRec and MIMN, focus on extracting different interests for a user by performing historical item clustering, while graph convolution methods including TGSRec and SURGE elect to refine user preferences based on multilevel correlations between historical items. Unfortunately, neither of them realizes that these two types of solutions can mutually complement each other, by aggregating multi-level user preference to achieve more precise multi-interest extraction for a better recommendation. To this end, in this paper, we propose a unified multi-grained neural model (named MGNM) via a combination of multi-interest learning and graph convolutional aggregation. Concretely, MGNM first learns the graph structure and information aggregation paths of the historical items for a user. It then performs graph convolution to derive item representations in an iterative fashion, in which the complex preferences at different levels can be well captured. Afterwards, a novel sequential capsule network is proposed to inject the sequential patterns into the multi-interest extraction process, leading to a more precise interest learning in a multi-grained manner. Experiments on three real-world datasets from different scenarios demonstrate the superiority of MGNM against several state-of-the-art baselines. The performance gain over the best baseline is up to 27.10% and 25.17% in terms of [email protected] and [email protected] respectively, which is one of the largest gains in recent development of sequential recommendation. Further analysis also demonstrates that MGNM is robust and effective at user preference understanding at multi-grained levels.

News recommendation is often modeled as a sequential recommendation task, assuming there are rich short-term dependencies over historical clicked news. However, users usually have strong preferences on the temporal diversity of news information and may not tend to click similar news successively, which is very different from many sequential recommendation scenarios such as e-commerce recommendation. In this paper, we study whether news recommendation can be regarded as a standard sequential recommendation problem. Through extensive experiments on two real-world datasets, we find it suboptimal to model news recommendation as a conventional sequential recommendation problem. To handle this issue, we further propose a temporal diversity-aware sequential news recommendation method that can promote candidate news that are diverse from recently clicked news to help predict future clicks more accurately. Experiments show that our method can empower various news recommendation methods.

Knowledge Graphs (KGs) have been utilized as useful side information to improve recommendation quality. In those recommender systems, knowledge graph information often contains fruitful facts and inherent semantic relatedness among items. However, the success of such methods relies on the high quality knowledge graphs, and may not learn quality representations with two challenges: i) The long-tail distribution of entities results in sparse supervision signals for KG-enhanced item representation; ii) Real-world knowledge graphs are often noisy and contain topic-irrelevant connections between items and entities. Such KG sparsity and noise make the item-entity dependent relations deviate from reflecting their true characteristics, which significantly amplifies the noise effect and hinders the accurate representation of user's preference.

Session-based recommendation aims to predict items that an anonymous user would like to purchase based on her short behavior sequence. The current approaches towards session-based recommendation only focus on modeling users' interest preferences, while they all ignore a key attribute of an item, i.e., the price. Many marketing studies have shown that the price factor significantly influences users' behaviors and the purchase decisions of users are determined by both price and interest preferences simultaneously. However, it is nontrivial to incorporate price preferences for session-based recommendation. Firstly, it is hard to handle heterogeneous information from various features of items to capture users' price preferences. Secondly, it is difficult to model the complex relations between price and interest preferences in determining user choices.

Multi-scenario learning (MSL) enables a service provider to cater for users' fine-grained demands by separating services for different user sectors, e.g., by user's geographical region. Under each scenario there is a need to optimize multiple task-specific targets e.g., click through rate and conversion rate, known as multi-task learning (MTL). Recent solutions for MSL and MTL are mostly based on the multi-gate mixture-of-experts (MMoE) architecture. MMoE structure is typically static and its design requires domain-specific knowledge, making it less effective in handling both MSL and MTL. In this paper, we propose a novel Automatic Expert Selection framework for Multi-scenario and Multi-task search, named AESM2. AESM2 integrates both MSL and MTL into a unified framework with an automatic structure learning. Specifically, AESM2 stacks multi-task layers over multi-scenario layers. This hierarchical design enables us to flexibly establish intrinsic connections between different scenarios, and at the same time also supports high-level feature extraction for different tasks. At each multi-scenario/multi-task layer, a novel expert selection algorithm is proposed to automatically identify scenario-/task-specific and shared experts for each input. Experiments over two real-world large-scale datasets demonstrate the effectiveness of AESM2 over a battery of strong baselines. Online A/B test also shows substantial performance gain on multiple metrics. Currently, AESM2 has been deployed online for serving major traffic.

Recent works have shown the effectiveness of incorporating textual and visual information to tackle the sparsity problem in recommendation scenarios. To fuse these useful heterogeneous modality information, an essential prerequisite is to align these information for modality-robust features learning and semantic understanding. Unfortunately, existing works mainly focus on tackling the learning of common knowledge across modalities, while the specific characteristics of each modality is discarded, which may inevitably degrade the recommendation performance.

Named entity recognition (NER) is the task to identify mentions of rigid designators from text belonging to predefined semantic types such as person, location, organization etc. NER always serves as the foundation for many natural language applications such as question answering, text summarization, and machine translation. Early NER systems got a huge success in achieving good performance with the cost of human engineering in designing domain-specific features and rules. In recent years, deep learning, empowered by continuous real-valued vector representations and semantic composition through nonlinear processing, has been employed in NER systems, yielding stat-of-the-art performance. In this paper, we provide a comprehensive review on existing deep learning techniques for NER. We first introduce NER resources, including tagged NER corpora and off-the-shelf NER tools. Then, we systematically categorize existing works based on a taxonomy along three axes: distributed representations for input, context encoder, and tag decoder. Next, we survey the most representative methods for recent applied techniques of deep learning in new NER problem settings and applications. Finally, we present readers with the challenges faced by NER systems and outline future directions in this area.

Recently, micro-video sharing platforms such as Kuaishou and Tiktok have become a major source of information for people's lives. Thanks to the large traffic volume, short video lifespan and streaming fashion of these services, it has become more and more pressing to improve the existing recommender systems to accommodate these challenges in a cost-effective way. In this paper, we propose a novel concept-aware denoising graph neural network (named Conde) for micro-video recommendation. Conde consists of a three-phase graph convolution process to derive user and micro-video representations: warm-up propagation, graph denoising and preference refinement. A heterogeneous tripartite graph is constructed by connecting user nodes with video nodes, and video nodes with associated concept nodes, extracted from captions and comments of the videos. To address the noisy information in the graph, we introduce a user-oriented graph denoising phase to extract a subgraph which can better reflect the user's preference. Despite the main focus of micro-video recommendation in this paper, we also show that our method can be generalized to other types of tasks. Therefore, we also conduct empirical studies on a well-known public E-commerce dataset. The experimental results suggest that the proposed Conde achieves significantly better recommendation performance than the existing state-of-the-art solutions.

Streaming spatial-textual data that contains geographic and textual information, e.g., geo-tagged tweets, has an unprecedented increase in amount. As one of the basic operations, the continuous spatial-textual queries that retrieve real-time results continuously on large-scale spatial-textual streams call for means of efficient distributed processing. However, existing proposals either are spatialaware only, or superficially exploit textual information for pruning. We propose a distributed system, called HASTE, for hybrid and adaptive processing on streaming spatial-textual data. The novelty lies on three aspects: (1) We propose a novel method to reduce the workload beforehand by dividing objects and queries into mutually exclusive types; (2) We develop a novel load partitioning strategy and a novel cost model that consider both spatial and textual properties; (3) We design a multi-level load adjustment strategy that adaptively copes with different degrees of load imbalance. We report on extensive experiments with real-world data that offer insight into the performance of the solution, and show that the solution is capable of outperforming the state-of-the-art proposals.

The large-scale recommender system mainly consists of two stages: matching and ranking. The matching stage (also known as the retrieval step) identifies a small fraction of relevant items from billion-scale item corpus in low latency and computational cost. Item-to-item collaborative filtering (item-based CF) and embedding-based retrieval (EBR) have been long used in the industrial matching stage owing to its efficiency. However, item-based CF is hard to meet personalization, while EBR has difficulty in satisfying diversity. In this paper, we propose a novel matching architecture, Path-based Deep Network (named PDN), through incorporating both personalization and diversity to enhance matching performance. Specifically, PDN is comprised of two modules: Trigger Net and Similarity Net. PDN utilizes Trigger Net to capture the user's interest in each of his/her interacted item. Similarity Net is devised to evaluate the similarity between each interacted item and the target item based on these items' profile and CF information. The final relevance between the user and the target item is calculated by explicitly considering user's diverse interests, \ie aggregating the relevance weights of the related two-hop paths (one hop of a path corresponds to user-item interaction and the other to item-item relevance). Furthermore, we describe the architecture design of the proposed PDN in a leading real-world E-Commerce service (Mobile Taobao App). Based on offline evaluations and online A/B test, we show that PDN outperforms the existing solutions for the same task. The online results also demonstrate that PDN can retrieve more personalized and more diverse items to significantly improve user engagement. Currently, PDN system has been successfully deployed at Mobile Taobao App and handling major online traffic.

In recent years, legal case retrieval has attracted much attention in the IR research community. It aims to retrieve supporting cases for a given query case and contributes to better legal systems. While using a legal case retrieval system, users always feel difficult to construct accurate queries to express their information need, especially when they lack sufficient domain knowledge. Since conversational search has been widely recognized to fulfill users' complex and exploratory information need, we investigate whether conversational search paradigm can be adopted to improve users' legal case retrieval experience. We design a laboratory-based study to collect users' interaction behaviors and explicit feedback signals while using traditional and agent-mediated conversational legal case retrieval systems. Based on the collected data, we compare search behavior and outcome of these two different kinds of interaction paradigms. Compared with the traditional one, experimental results show that users can achieve better retrieval performance with the conversational case retrieval system. Moreover, conversational system can also save users' efforts in formulating queries and examining results.

Next basket recommendation aims to infer a set of items that a user will purchase at the next visit by considering a sequence of baskets he/she has purchased previously. This task has drawn increasing attention from both the academic and industrial communities. The existing solutions mainly focus on sequential modeling over their historical interactions. However, due to the diversity and randomness of users' behaviors, not all these baskets are relevant to help identify the user's next move. It is necessary to denoise the baskets and extract credibly relevant items to enhance recommendation performance. Unfortunately, this dimension is usually overlooked in the current literature.

Recently, exploiting a knowledge graph (KG) to enrich the semantic representation of a news article have been proven to be effective for news recommendation. These solutions focus on the representation learning for news articles with additional information in the knowledge graph, where the user representations are mainly derived based on these news representations later. However, different users would hold different interests on the same news article. In other words, directly identifying the entities relevant to the user's interest and deriving the resultant user representation could enable a better news recommendation and explanation.

Sharing recommendation is becoming ubiquitous at almost every e-commerce website, where a user will be recommended a list of users when he wants to share something with others. With the tremendous growth of online shopping users, sharing recommendation confronts several distinct difficulties: 1) how to establish a unified recommender model for large numbers of sharing scenarios; 2) how to handle with long-tail even cold start scenarios with limited training data; 3) how to incorporate social influence in order to make more accurate recommendations.

In e-commerce search, vectorized matching is the most important approach besides lexical matching, where learning vector representations for entities (e.g., query, item, shop) plays a crucial role. In this work, we focus on vectorized search matching model for shop search in Taobao. Unlike item search, shop search is faced with serious behavior sparsity and long-tail problem. To tackle this, we take the first step to transfer knowledge from item search, i.e., leveraging items purchased under a query and the shops they belong to. Moreover, we propose a novel gated heterogeneous graph learning model (named GHL) to derive vector representations for entities. Both first-order and second-order proximity of queries and shops are exploited to fully mine the heterogeneous relationships. And to relieve long-tail phenomenon, we devise an innovative gated neighbor aggregation scheme where each type of entities (i.e., hot ones and long-tail ones) can benefit from the heterogeneous graph in an automatic way. Finally, the whole framework is jointly trained in an end-to-end fashion. Offline evaluation results on real-world data of Taobao shop search platform demonstrate that the proposed model outperforms existing graph based methods, and online A/B tests show that it is highly effective and achieves significant CTR improvements.

Shop search has become an increasingly important service provided by Taobao, the China's largest e-commerce platform. By using shop search, a user can easily identify the desired shop that provides a full-scale of relevant items matching his information need. With the tremendous growth of users and shops, shop search faces several unique challenging problems: 1) many shop names do not fully express what they sell, i.e., the semantic gap between user query and shop name; 2) due to the lack of user interactions, it is difficult to deliver a good search result for the long-tail queries and retrieve long-tail shops that are highly relevant to a query.

Most of ranking models are trained only with displayed items (most are hot items), but they are utilized to retrieve items in the entire space which consists of both displayed and non-displayed items (most are long-tail items). Due to the sample selection bias, the long-tail items lack sufficient records to learn good feature representations, ie data sparsity and cold start problems. The resultant distribution discrepancy between displayed and non-displayed items would cause poor long-tail performance. To this end, we propose an entire space adaptation model (ESAM) to address this problem from the perspective of domain adaptation (DA). ESAM regards displayed and non-displayed items as source and target domains respectively. Specifically, we design the attribute correlation alignment that considers the correlation between high-level attributes of the item to achieve distribution alignment. Furthermore, we introduce two effective regularization strategies, ie center-wise clustering andself-training to improve DA process. Without requiring any auxiliary information and auxiliary domains, ESAM transfers the knowledge from displayed items to non-displayed items for alleviating the distribution inconsistency. Experiments on two public datasets and a large-scale industrial dataset collected from Taobao demonstrate that ESAM achieves state-of-the-art performance, especially in the long-tail space. Besides, we deploy ESAM to the Taobao search engine, leading to significant improvement on online performance. The code is available at https://github.com/A-bone1/ESAM.git.

Popularity is often included in experimental evaluation to provide areference performance for a recommendation task. To understand how popularity baseline is defined and evaluated, we sample 12 papers from top-tier conferences including KDD, WWW, SIGIR, and RecSys, and 6 open source toolkits. We note that the widely adoptedMostPop baseline simply ranks items based on the number of interactions in the training data. We argue that the current evaluation of popularity (i) does not reflect the popular items at the time when a user interacts with the system, and (ii) may recommend items released after a user's last interaction with the system. On the widely used MovieLens dataset, we show that the performance of popularity could be significantly improved by 70% or more, if we consider the popular itemsat the time point when a user interacts with the system. We further show that, on MovieLens dataset, the users having lower tendencies on movies tend to follow the crowd and rate more popular movies. Movie lovers who rate a large number of movies, rate movies based on their own preferences and interests. Through this study, we call for a re-visit of the popularity baseline in recommender system to better reflect its effectiveness.

In a large recommender system, the products (or items) could be in many different categories or domains. Given two relevant domains (e.g., Book and Movie), users may have interactions with items in one domain but not in the other domain. To the latter, these users are considered as cold-start users. How to effectively transfer users' preferences based on their interactions from one domain to the other relevant domain, is the key issue in cross-domain recommendation. Inspired by the advances made in review-based recommendation, we propose to model user preference transfer at aspect-level derived from reviews. To this end, we propose a cross-domain recommendation framework via aspect transfer network for cold-start users (named CATN). CATN is devised to extract multiple aspects for each user and each item from their review documents, and learn aspect correlations across domains with an attention mechanism. In addition, we further exploit auxiliary reviews from like-minded users to enhance a user's aspect representations. Then, an end-to-end optimization framework is utilized to strengthen the robustness of our model. On real-world datasets, the proposed CATN outperforms SOTA models significantly in terms of rating prediction accuracy. Further analysis shows that our model is able to reveal user aspect connections across domains at a fine level of granularity, making the recommendation explainable.

With the development of high computational devices, deep neural networks (DNNs), in recent years, have gained significant popularity in many Artificial Intelligence (AI) applications. However, previous efforts have shown that DNNs are vulnerable to strategically modified samples, named adversarial examples . These samples are generated with some imperceptible perturbations, but can fool the DNNs to give false predictions. Inspired by the popularity of generating adversarial examples against DNNs in Computer Vision (CV), research efforts on attacking DNNs for Natural Language Processing (NLP) applications have emerged in recent years. However, the intrinsic difference between image (CV) and text (NLP) renders challenges to directly apply attacking methods in CV to NLP. Various methods are proposed addressing this difference and attack a wide range of NLP applications. In this article, we present a systematic survey on these works. We collect all related academic works since the first appearance in 2017. We then select, summarize, discuss, and analyze 40 representative works in a comprehensive way. To make the article self-contained, we cover preliminary knowledge of NLP and discuss related seminal works in computer vision. We conclude our survey with a discussion on open issues to bridge the gap between the existing progress and more robust adversarial attacks on NLP DNNs.

With the large quantity of textual information produced on the Internet, a critical necessity is to filter out the irrelevant information and organize the rest into categories of interest (e.g., an emerging event). However, supervised-learning document filtering methods heavily rely on a large number of labeled documents for model training. Manually identifying plenty of positive examples for each category is expensive and time-consuming. Also, it is unrealistic to cover all the categories from an evolving text source that covers diverse kinds of events, user opinions, and daily life activities. In this article, we propose a novel attention-based deep relevance model for few-shot document filtering (named ADRM), inspired by the relevance feedback methodology proposed for ad hoc retrieval. ADRM calculates the relevance score between a document and a category by taking a set of seed words and a few seed documents relevant to the category. It constructs the category-specific conceptual representation of the document based on the corresponding seed words and seed documents. Specifically, to filter irrelevant yet noisy information in the seed documents, ADRM employs two types of attention mechanisms (namely whole-match attention and max-match attention ) and generates category-specific representations for them. Then ADRM is devised to extract the relevance signals by modeling the hidden feature interactions in the word embedding space. The relevance signals are extracted through a gated convolutional process, a self-attention layer, and a relevance aggregation layer. Extensive experiments on three real-world datasets show that ADRM consistently outperforms the existing technical alternatives, including the conventional classification and retrieval baselines, and the state-of-the-art deep relevance ranking models for few-shot document filtering. We also perform an ablation study to demonstrate that each component in ADRM is effective for enhancing filtering performance. Further analysis shows that ADRM is robust under varying parameter settings.

In recent years, with the prevalence of social media and smart devices, people causally reveal their locations such as shops, hotels, and restaurants in their tweets. Recognizing and linking such fine-grained location mentions to well-defined location profiles are beneficial for retrieval and recommendation systems. In this paper, we propose DLocRL, a new deep learning pipeline for fine-grained location recognition and linking in tweets, and verify its effectiveness on a real-world Twitter dataset.

Collective entity disambiguation, or collective entity linking aims to jointly resolve multiple mentions by linking them to their associated entities in a knowledge base. Previous works are primarily based on the underlying assumption that entities within the same document are highly related. However, the extent to which these entities are actually connected in reality is rarely studied and therefore raises interesting research questions. For the first time, this paper shows that the semantic relationships between mentioned entities within a document are in fact less dense than expected. This could be attributed to several reasons such as noise, data sparsity, and knowledge base incompleteness. As a remedy, we introduce MINTREE, a new tree-based objective for the problem of entity disambiguation. The key intuition behind MINTREE is the concept of coherence relaxation which utilizes the weight of a minimum spanning tree to measure the coherence between entities. Based on this new objective, we design Pair-Linking, a novel iterative solution for the MINTREE optimization problem. The idea of Pair-Linking is simple: instead of considering all the given mentions, Pair-Linking iteratively selects a pair with the highest confidence at each step for decision making. Via extensive experiments on eight benchmark datasets, we show that our approach is not only more accurate but also surprisingly faster than many state-of-the-art collective linking algorithms.

One important necessity is to filter out the irrelevant information and organize the relevant information into meaningful categories. However, developing text classifiers often requires a large number of labeled documents as training examples. Manually labeling documents is costly and time-consuming. More importantly, it becomes unrealistic to know all the categories covered by the documents beforehand. Recently, a few methods have been proposed to label documents by using a small set of relevant keywords for each category, known as dataless text classification . In this article, we propose a seed-guided topic model for the dataless text filtering and classification (named DFC). Given a collection of unlabeled documents, and for each specified category a small set of seed words that are relevant to the semantic meaning of the category, DFC filters out the irrelevant documents and classifies the relevant documents into the corresponding categories through topic influence. DFC models two kinds of topics: category-topics and general-topics . Also, there are two kinds of category-topics: relevant-topics and irrelevant-topics. Each relevant-topic is associated with one specific category, representing its semantic meaning. The irrelevant-topics represent the semantics of the unknown categories covered by the document collection. And the general-topics capture the global semantic information. DFC assumes that each document is associated with a single category-topic and a mixture of general-topics. A novelty of the model is that DFC learns the topics by exploiting the explicit word co-occurrence patterns between the seed words and regular words (i.e., non-seed words) in the document collection. A document is then filtered, or classified, based on its posterior category-topic assignment. Experiments on two widely used datasets show that DFC consistently outperforms the state-of-the-art dataless text classifiers for both classification with filtering and classification without filtering. In many tasks, DFC can also achieve comparable or even better classification accuracy than the state-of-the-art supervised learning solutions. Our experimental results further show that DFC is insensitive to the tuning parameters. Moreover, we conduct a thorough study about the impact of seed words for existing dataless text classification techniques. The results reveal that it is not using more seed words but the document coverage of the seed words for the corresponding category that affects the dataless classification performance.

Both reviews and user-item interactions (i.e., rating scores) have been widely adopted for user rating prediction. However, these existing techniques mainly extract the latent representations for users and items in an independent and static manner. That is, a single static feature vector is derived to encode user preference without considering the particular characteristics of each candidate item. We argue that this static encoding scheme is incapable of fully capturing users’ preferences, because users usually exhibit different preferences when interacting with different items. In this article, we propose a novel c ontext- a ware user-item r epresentation l earning model for rating prediction, named CARL. CARL derives a joint representation for a given user-item pair based on their individual latent features and latent feature interactions. Then, CARL adopts Factorization Machines to further model higher order feature interactions on the basis of the user-item pair for rating prediction. Specifically, two separate learning components are devised in CARL to exploit review data and interaction data, respectively: review-based feature learning and interaction-based feature learning . In the review-based learning component, with convolution operations and attention mechanism, the pair-based relevant features for the given user-item pair are extracted by jointly considering their corresponding reviews. However, these features are only reivew-driven and may not be comprehensive. Hence, an interaction-based learning component further extracts complementary features from interaction data alone, also on the basis of user-item pairs. The final rating score is then derived with a dynamic linear fusion mechanism. Experiments on seven real-world datasets show that CARL achieves significantly better rating prediction accuracy than existing state-of-the-art alternatives. Also, with the attention mechanism, we show that the pair-based relevant information (i.e., context-aware information) in reviews can be highlighted to interpret the rating prediction for different user-item pairs.

Review-based methods are one of the dominant methods to address the data sparsity problem of recommender system. However, the performance of most existing review-based methods will degrade when the review is also sparse. To this end, we propose a method to exploit user-item p air-dependent features from a uxiliary r eviews written by l ike-minded users (PARL) to address such problem. That is, both the reviews written by the user and the reviews written for the item are incorporated to highlight the useful features covered by the auxiliary reviews. PARL not only alleviates the sparsity problem of reviews but also produce extra informative features to further improve the accuracy of rating prediction. More importantly, it is designed as a plug-and-play model which can be plugged into various deep recommender systems to improve recommendations provided by them. Extensive experiments on five real-world datasets show that PARL achieves better prediction accuracy than other state-of-the-art alternatives. Also, with the exploitation of auxiliary reviews, the performance of PARL is robust on datasets with different characteristics.

Entity disambiguation, also known as entity linking, is the task of mapping mentions in text to the corresponding entities in a given knowledge base, e.g. Wikipedia. Two key challenges are making use of mention's context to disambiguate (i.e. local objective), and promoting coherence of all the linked entities (i.e. global objective). In this paper, we propose a deep neural network model to effectively measure the semantic matching between mention's context and target entity. We are the first to employ the long short-term memory (LSTM) and attention mechanism for entity disambiguation. We also propose Pair-Linking, a simple but effective and significantly fast linking algorithm. Pair-Linking iteratively identifies and resolves pairs of mentions, starting from the most confident pair. It finishes linking all mentions in a document by scanning the pairs of mentions at most once. Our neural network model combined with Pair-Linking, named NeuPL, outperforms state-of-the-art systems over different types of documents including news, RSS, and tweets.

Twitter has attracted billions of users for life logging and sharing activities and opinions. In their tweets, users often reveal their location information and short‐term visiting histories or plans. Capturing user's short‐term activities could benefit many applications for providing the right context at the right time and location. In this paper we are interested in extracting locations mentioned in tweets at fine‐grained granularity, with temporal awareness. Specifically, we recognize the points‐of‐interest (POIs) mentioned in a tweet and predict whether the user has visited, is currently at, or will soon visit the mentioned POIs. A POI can be a restaurant, a shopping mall, a bookstore, or any other fine‐grained location. Our proposed framework, named TS‐Petar (Two‐Stage POI Extractor with Temporal Awareness), consists of two main components: a POI inventory and a two‐stage time‐aware POI tagger. The POI inventory is built by exploiting the crowd wisdom of the Foursquare community. It contains both POIs' formal names and their informal abbreviations, commonly observed in Foursquare check‐ins. The time‐aware POI tagger, based on the Conditional Random Field (CRF) model, is devised to disambiguate the POI mentions and to resolve their associated temporal awareness accordingly. Three sets of contextual features (linguistic, temporal, and inventory features) and two labeling schema features (OP and BILOU schemas) are explored for the time‐aware POI extraction task. Our empirical study shows that the subtask of POI disambiguation and the subtask of temporal awareness resolution call for different feature settings for best performance. We have also evaluated the proposed TS‐Petar against several strong baseline methods. The experimental results demonstrate that the two‐stage approach achieves the best accuracy and outperforms all baseline methods in terms of both effectiveness and efficiency.

Many applications require semantic understanding of short texts, and inferring discriminative and coherent latent topics is a critical and fundamental task in these applications. Conventional topic models largely rely on word co-occurrences to derive topics from a collection of documents. However, due to the length of each document, short texts are much more sparse in terms of word co-occurrences. Recent studies show that the Dirichlet Multinomial Mixture (DMM) model is effective for topic inference over short texts by assuming that each piece of short text is generated by a single topic. However, DMM has two main limitations. First, even though it seems reasonable to assume that each short text has only one topic because of its shortness, the definition of “shortness” is subjective and the length of the short texts is dataset dependent. That is, the single-topic assumption may be too strong for some datasets. To address this limitation, we propose to model the topic number as a Poisson distribution, allowing each short text to be associated with a small number of topics (e.g., one to three topics). This model is named PDMM. Second, DMM (and also PDMM) does not have access to background knowledge (e.g., semantic relations between words) when modeling short texts. When a human being interprets a piece of short text, the understanding is not solely based on its content words, but also their semantic relations. Recent advances in word embeddings offer effective learning of word semantic relations from a large corpus. Such auxiliary word embeddings enable us to address the second limitation. To this end, we propose to promote the semantically related words under the same topic during the sampling process, by using the generalized Pólya urn (GPU) model. Through the GPU model, background knowledge about word semantic relations learned from millions of external documents can be easily exploited to improve topic modeling for short texts. By directly extending the PDMM model with the GPU model, we propose two more effective topic models for short texts, named GPU-DMM and GPU-PDMM. Through extensive experiments on two real-world short text collections in two languages, we demonstrate that PDMM achieves better topic representations than state-of-the-art models, measured by topic coherence. The learned topic representation leads to better accuracy in a text classification task, as an indirect evaluation. Both GPU-DMM and GPU-PDMM further improve topic coherence and text classification accuracy. GPU-PDMM outperforms GPU-DMM at the price of higher computational costs.

Developing text classifiers often requires a large number of labeled documents as training examples. However, manually labeling documents is costly and time-consuming. Recently, a few methods have been proposed to label documents by using a small set of relevant keywords for each category, known as dataless text classification. In this paper, we propose a Seed-Guided Topic Model (named STM) for the dataless text classification task. Given a collection of unlabeled documents, and for each category a small set of seed words that are relevant to the semantic meaning of the category, the STM predicts the category labels of the documents through topic influence. STM models two kinds of topics: category-topics and general-topics. Each category-topic is associated with one specific category, representing its semantic meaning. The general-topics capture the global semantic information underlying the whole document collection. STM assumes that each document is associated with a single category-topic and a mixture of general-topics. A novelty of the model is that STM learns the topics by exploiting the explicit word co-occurrence patterns between the seed words and regular words (i.e., non-seed words) in the document collection. A document is then labeled, or classified, based on its posterior category-topic assignment. Experiments on two widely used datasets show that STM consistently outperforms the state-of-the-art dataless text classifiers. In some tasks, STM can also achieve comparable or even better classification accuracy than the state-of-the-art supervised learning solutions. Our experimental results further show that STM is insensitive to the tuning parameters. Stable performance with little variation can be achieved in a broad range of parameter settings, making it a desired choice for real applications.

For many applications that require semantic understanding of short texts, inferring discriminative and coherent latent topics from short texts is a critical and fundamental task. Conventional topic models largely rely on word co-occurrences to derive topics from a collection of documents. However, due to the length of each document, short texts are much more sparse in terms of word co-occurrences. Data sparsity therefore becomes a bottleneck for conventional topic models to achieve good results on short texts. On the other hand, when a human being interprets a piece of short text, the understanding is not solely based on its content words, but also her background knowledge (e.g., semantically related words). The recent advances in word embedding offer effective learning of word semantic relations from a large corpus. Exploiting such auxiliary word embeddings to enrich topic modeling for short texts is the main focus of this paper. To this end, we propose a simple, fast, and effective topic model for short texts, named GPU-DMM. Based on the Dirichlet Multinomial Mixture (DMM) model, GPU-DMM promotes the semantically related words under the same topic during the sampling process by using the generalized Polya urn (GPU) model. In this sense, the background knowledge about word semantic relatedness learned from millions of external documents can be easily exploited to improve topic modeling for short texts. Through extensive experiments on two real-world short text collections in two languages, we show that GPU-DMM achieves comparable or better topic representations than state-of-the-art models, measured by topic coherence. The learned topic representation leads to the best accuracy in text classification task, which is used as an indirect evaluation.

Twitter has attracted millions of users to share and disseminate most up-to-date information, resulting in large volumes of data produced everyday. However, many applications in Information Retrieval (IR) and Natural Language Processing (NLP) suffer severely from the noisy and short nature of tweets. In this paper, we propose a novel framework for tweet segmentation in a batch mode, called HybridSeg. By splitting tweets into meaningful segments, the semantic or context information is well preserved and easily extracted by the downstream applications. HybridSeg finds the optimal segmentation of a tweet by maximizing the sum of the stickiness scores of its candidate segments. The stickiness score considers the probability of a segment being a phrase in English (i.e., global context) and the probability of a segment being a phrase within the batch of tweets (i.e., local context). For the latter, we propose and evaluate two models to derive local context by considering the linguistic features and term-dependency in a batch of tweets, respectively. HybridSeg is also designed to iteratively learn from confident segments as pseudo feedback. Experiments on two tweet data sets show that tweet segmentation quality is significantly improved by learning both global and local contexts compared with using global context alone. Through analysis and comparison, we show that local linguistic features are more reliable for learning local context compared with term-dependency. As an application, we show that high accuracy is achieved in named entity recognition by applying segment-based part-of-speech (POS) tagging.

Twitter is a popular platform for sharing activities, plans, and opinions. Through tweets, users often reveal their location information and short term visiting plans. In this paper, we are interested in extracting fine-grained locations mentioned in tweets with temporal awareness. More specifically, we like to extract each point-of-interest (POI) mention in a tweet and predict whether the user has visited, is currently at, or will soon visit this POI. Our proposed solution, named PETAR, consists of two main components: a POI inventory and a time-aware POI tagger. The POI inventory is built by exploiting the crowd wisdom of Foursquare community. It contains not only the formal names of POIs but also the informal abbreviations. The POI tagger, based on Conditional Random Field (CRF) model, is designed to simultaneously identify the POIs and resolve their associated temporal awareness. In our experiments, we investigated four types of features (i.e., lexical, grammatical, geographical, and BILOU schema features) for time-aware POI extraction. With the four types of features, PETAR achieves promising extraction accuracy and outperforms all baseline methods.

Twitter has attracted hundred millions of users to share and disseminate most up-to-date information. However, the noisy and short nature of tweets makes many applications in information retrieval (IR) and natural language processing (NLP) challenging. Recently, segment-based tweet representation has demonstrated effectiveness in named entity recognition (NER) and event detection from tweet streams. To split tweets into meaningful phrases or segments, the previous work is purely based on external knowledge bases, which ignores the rich local context information embedded in the tweets. In this paper, we propose a novel framework for tweet segmentation in a batch mode, called HybridSeg. HybridSeg incorporates local context knowledge with global knowledge bases for better tweet segmentation. HybridSeg consists of two steps: learning from off-the-shelf weak NERs and learning from pseudo feedback. In the first step, the existing NER tools are applied to a batch of tweets. The named entities recognized by these NERs are then employed to guide the tweet segmentation process. In the second step, HybridSeg adjusts the tweet segmentation results iteratively by exploiting all segments in the batch of tweets in a collective manner. Experiments on two tweet datasets show that HybridSeg significantly improves tweet segmentation quality compared with the state-of-the-art algorithm. We also conduct a case study by using tweet segments for the task of named entity recognition from tweets. The experimental results demonstrate that HybridSeg significantly benefits the downstream applications.

The semantic knowledge of Wikipedia has proved to be useful for many tasks, for example, named entity disambiguation. Among these applications, the task of identifying the word sense based on Wikipedia is a crucial component because the output of this component is often used in subsequent tasks. In this article, we present a two‐stage framework (called TSDW) for word sense disambiguation using knowledge latent in Wikipedia. The disambiguation of a given phrase is applied through a two‐stage disambiguation process: (a) The first‐stage disambiguation explores the contextual semantic information, where the noisy information is pruned for better effectiveness and efficiency; and (b) the second‐stage disambiguation explores the disambiguated phrases of high confidence from the first stage to achieve better redisambiguation decisions for the phrases that are difficult to disambiguate in the first stage. Moreover, existing studies have addressed the disambiguation problem for English text only. Considering the popular usage of Wikipedia in different languages, we study the performance of TSDW and the existing state‐of‐the‐art approaches over both English and Traditional Chinese articles. The experimental results show that TSDW generalizes well to different semantic relatedness measures and text in different languages. More important, TSDW significantly outperforms the state‐of‐the‐art approaches with both better effectiveness and efficiency.

Event detection from tweets is an important task to understand the current events/topics attracting a large number of common users. However, the unique characteristics of tweets (e.g. short and noisy content, diverse and fast changing topics, and large data volume) make event detection a challenging task. Most existing techniques proposed for well written documents (e.g. news articles) cannot be directly adopted. In this paper, we propose a segment-based event detection system for tweets, called Twevent. Twevent first detects bursty tweet segments as event segments and then clusters the event segments into events considering both their frequency distribution and content similarity. More specifically, each tweet is split into non-overlapping segments (i.e. phrases possibly refer to named entities or semantically meaningful information units). The bursty segments are identified within a fixed time window based on their frequency patterns, and each bursty segment is described by the set of tweets containing the segment published within that time window. The similarity between a pair of bursty segments is computed using their associated tweets. After clustering bursty segments into candidate events, Wikipedia is exploited to identify the realistic events and to derive the most newsworthy segments to describe the identified events. We evaluate Twevent and compare it with the state-of-the-art method using 4.3 million tweets published by Singapore-based users in June 2010. In our experiments, Twevent outperforms the state-of-the-art method by a large margin in terms of both precision and recall. More importantly, the events detected by Twevent can be easily interpreted with little background knowledge because of the newsworthy segments. We also show that Twevent is efficient and scalable, leading to a desirable solution for event detection from tweets.

Many private and/or public organizations have been reported to create and monitor targeted Twitter streams to collect and understand users' opinions about the organizations. Targeted Twitter stream is usually constructed by filtering tweets with user-defined selection criteria e.g. tweets published by users from a selected region, or tweets that match one or more predefined keywords. Targeted Twitter stream is then monitored to collect and understand users' opinions about the organizations. There is an emerging need for early crisis detection and response with such target stream. Such applications require a good named entity recognition (NER) system for Twitter, which is able to automatically discover emerging named entities that is potentially linked to the crisis. In this paper, we present a novel 2-step unsupervised NER system for targeted Twitter stream, called TwiNER. In the first step, it leverages on the global context obtained from Wikipedia and Web N-Gram corpus to partition tweets into valid segments (phrases) using a dynamic programming algorithm. Each such tweet segment is a candidate named entity. It is observed that the named entities in the targeted stream usually exhibit a gregarious property, due to the way the targeted stream is constructed. In the second step, TwiNER constructs a random walk model to exploit the gregarious property in the local context derived from the Twitter stream. The highly-ranked segments have a higher chance of being true named entities. We evaluated TwiNER on two sets of real-life tweets simulating two targeted streams. Evaluated using labeled ground truth, TwiNER achieves comparable performance as with conventional approaches in both streams. Various settings of TwiNER have also been examined to verify our global context + local context combo idea.

The common tags given by multiple users to a particular document are often semantically relevant to the document and each tag represents a specific topic. In this paper, we attempt to emulate human tagging behavior to recommend tags by considering the concepts contained in documents. Specifically, we represent each document using a few most relevant concepts contained in the document, where the concept space is derived from Wikipedia. Tags are then recommended based on the tag concept model derived from the annotated documents of each tag. Evaluated on a Delicious dataset of more than 53K documents, the proposed technique achieved comparable tag recommendation accuracy as the state-of-the-art, while yielding an order of magnitude speed-up.