Large Language Models (LLMs) have shown strong capabilities in open-domain question answering (QA), but deploying them in real-world online systems introduces critical challenges. These include: (1) handling both simple and complex queries with appropriate levels of reasoning, (2) minimizing latency without compromising answer quality, and (3) maintaining answer consistency under evolving and noisy retrieval contexts. To address these challenges, we propose Retain-to-Refine (ℜ2ℜ), an adaptive agent-based QA framework designed for practical deployment. ℜ2ℜ integrates a Query Critic Agent (QCA) to assess query difficulty and route it accordingly: simple queries are answered directly using fast, prompt-based LLM calls, while complex queries are handled by a Memory Augmented Agent (MAA). MAA performs iterative reasoning guided by a unique long-short memory mechanism. Long-term memory retains and consolidates stable, core facts to ground the reasoning process, while short-term memory identifies transient information gaps to formulate highly focused subsequent queries. To ensure evidence quality, a Supervised Retrospection module validates and filters retrieved documents at each step. This agent-based design enables ℜ2ℜ to dynamically allocate computation based on question complexity, reducing unnecessary overhead while preserving high-quality answers when multi-step reasoning or external knowledge is required. Extensive evaluations across various settings and datasets demonstrate that the efficiency of R2R across diverse question types. In online settings, ℜ2ℜ delivers substantial gains in both response quality and efficiency, making it well-suited for large-scale industrial deployment in real-time QA services.

Generative inference with Large Language Models (LLMs) is the cornerstone of web-scale AI search, where queries are answered using vast, heterogeneous documents retrieved via Retrieval-Augmented Generation (RAG). This paradigm is critically bottlenecked by the cost of self-attention mechanism on long context. The sheer diversity of retrieved web content (multi-sourced, multi-lingual, multi-faceted) makes simple Key-Value (KV) cache optimizations with pre-fixed subsets ineffective, demanding a dynamic, content-aware approach. This challenge, however, introduces a classic chicken-and-egg problem: the model cannot foresee the necessary KV entries for attention without first inferring on the content, yet doing so on the full context is prohibitively expensive. This paper introduces P&F, a unified framework that resolves this dilemma through a core ''probe-and-fetch'' mechanism, which ingeniously integrates with speculative decoding -- an acceleration approach already adopted in web-scale AI search. The probe step repurposes the speculative draft model: while generating candidate tokens, it simultaneously probes the context to predict the most salient KV entries the large model will need for attention. The fetch step immediately acts on this prediction, asynchronously fetching these sparse entries from memory. This synergistic design piggybacks the probing step onto the drafting process, allowing the expensive gathering of a sparse KV cache to be fully masked. Crucially, this co-design breaks the sequential dependency bottleneck that cripples naive integrations of speculative decoding and prefetching due to synchronization issues. Extensive experiments show P&F significantly outperforms state-of-the-art methods in throughput and scalability, offering a practical, drop-in solution. Extensive offline evaluations across various settings and datasets demonstrate that P&F yields superior throughput and scalability compared to advanced baselines, while maintaining model quality across diverse models and scales. In online settings, P&F delivers substantial gains in throughput improvements while preserving response quality, making it well-suited for large-scale industrial deployment in real-time AI Search services.

Recent advances in deep learning have significantly boosted performance in multivariate time series forecasting (MTSF). While many existing approaches focus on capturing inter-variable (a.k.a. channel-wise) correlations to improve prediction accuracy, the temporal dimension, particularly its rich structural and contextual information, remains underexplored. In this paper, we propose BIM3, a novel framework that integrates BIdirectional temporal-aware modeling with Multi-Scale Mixture-of-Experts for MTSF. First, unlike existing methods that treat historical and future temporal information independently, we introduce a novel Timestamp Dual Cross-Attention Module, which employs a symmetric cross-attention mechanism to explicitly capture bidirectional temporal dependencies through timestamp interactions. Second, to address the complex and scale-varying temporal patterns commonly found in multivariate time series, we move beyond recent multi-scale forecasting models that share parameters across all channels and fail to capture channel-specific dynamics. Instead, we design a Multi-Scale Feature Extract Mixture-of-Experts module that adaptively routes time series to specialized experts based on their temporal characteristics. Extensive experiments on multiple real-world datasets show that BIM3 consistently outperforms state-of-the-art methods, highlighting its effectiveness in capturing both temporal structure and inter-variable diversity.

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.

Text-video retrieval is a challenging task that aims to identify relevant videos given textual queries. Compared to conventional textual retrieval, the main obstacle for text-video retrieval is the semantic gap between the textual nature of queries and the visual richness of video content. Previous works primarily focus on aligning the query and the video by finely aggregating word-frame matching signals. Inspired by the human cognitive process of modularly judging the relevance between text and video, the judgment needs high-order matching signal due to the consecutive and complex nature of video contents. In this paper, we propose chunk-level text-video matching, where the query chunks are extracted to describe a specific retrieval unit, and the video chunks are segmented into distinct clips from videos. We formulate the chunk-level matching as n-ary correlations modeling between words of the query and frames of the video and introduce a multi-modal hypergraph for n-ary correlation modeling. By representing textual units and video frames as nodes and using hyperedges to depict their relationships, a multi-modal hypergraph is constructed. In this way, the query and the video can be aligned in a high-order semantic space. In addition, to enhance the model's generalization ability, the extracted features are fed into a variational inference component for computation, obtaining the variational representation under the Gaussian distribution. The incorporation of hypergraphs and variational inference allows our model to capture complex, n-ary interactions among textual and visual contents. Experimental results demonstrate that our proposed method achieves state-of-the-art performance on the text-video retrieval task.

The page presentation biases in the information retrieval system, especially on the click behavior, is a well-known challenge that hinders improving ranking models' performance with implicit user feedback. Unbiased Learning to Rank~(ULTR) algorithms are then proposed to learn an unbiased ranking model with biased click data. However, most existing algorithms are specifically designed to mitigate position-related bias, e.g., trust bias, without considering biases induced by other features in search result page presentation(SERP), e.g. attractive bias induced by the multimedia. Unfortunately, those biases widely exist in industrial systems and may lead to an unsatisfactory search experience. Therefore, we introduce a new problem, i.e., whole-page Unbiased Learning to Rank(WP-ULTR), aiming to handle biases induced by whole-page SERP features simultaneously. It presents tremendous challenges: (1) a suitable user behavior model (user behavior hypothesis) can be hard to find; and (2) complex biases cannot be handled by existing algorithms. To address the above challenges, we propose a Bias Agnostic whole-page unbiased Learning to rank algorithm, named BAL, to automatically find the user behavior model with causal discovery and mitigate the biases induced by multiple SERP features with no specific design. Experimental results on a real-world dataset verify the effectiveness of the BAL.

Recent studies show that graph neural networks (GNNs) are prevalent to model high-order relationships for collaborative filtering (CF). Towards this research line, graph contrastive learning (GCL) has exhibited powerful performance in addressing the supervision label shortage issue by learning augmented user and item representations. While many of them show their effectiveness, two key questions still remain unexplored: i) Most existing GCL-based CF models are still limited by ignoring the fact that user-item interaction behaviors are often driven by diverse latent intent factors (e.g., shopping for family party, preferred color or brand of products); ii) Their introduced non-adaptive augmentation techniques are vulnerable to noisy information, which raises concerns about the model's robustness and the risk of incorporating misleading self-supervised signals. In light of these limitations, we propose a Disentangled Contrastive Collaborative Filtering framework (DCCF) to realize intent disentanglement with self-supervised augmentation in an adaptive fashion. With the learned disentangled representations with global context, our DCCF is able to not only distill finer-grained latent factors from the entangled self-supervision signals but also alleviate the augmentation-induced noise. Finally, the cross-view contrastive learning task is introduced to enable adaptive augmentation with our parameterized interaction mask generator. Experiments on various public datasets demonstrate the superiority of our method compared to existing solutions. Our model implementation is released at the link https://github.com/HKUDS/DCCF.

The pre-trained language models (PLMs), such as BERT and ERNIE, have achieved outstanding performance in many natural language understanding tasks. Recently, PLMs-based Information Retrieval models have also been investigated and showed substantially state-of-the-art effectiveness, e.g., MORES, PROP and ColBERT. Moreover, most of the PLMs-based rankers only focus on a single level relevance matching (e.g., character-level), while ignore the other granularity information (e.g., words and phrases), which easily lead to the ambiguity of query understanding and inaccurate matching issues in web search.

Collaborative Filtering (CF) has emerged as fundamental paradigms for parameterizing users and items into latent representation space, with their correlative patterns from interaction data. Among various CF techniques, the development of GNN-based recommender systems, e.g., PinSage and LightGCN, has offered the state-of-the-art performance. However, two key challenges have not been well explored in existing solutions: i) The over-smoothing effect with deeper graph-based CF architecture, may cause the indistinguishable user representations and degradation of recommendation results. ii) The supervision signals (i.e., user-item interactions) are usually scarce and skewed distributed in reality, which limits the representation power of CF paradigms. To tackle these challenges, we propose a new self-supervised recommendation framework Hypergraph Contrastive Collaborative Filtering (HCCF) to jointly capture local and global collaborative relations with a hypergraph-enhanced cross-view contrastive learning architecture. In particular, the designed hypergraph structure learning enhances the discrimination ability of GNN-based CF paradigm, in comprehensively capturing the complex high-order dependencies among users. Additionally, our HCCF model effectively integrates the hypergraph structure encoding with self-supervised learning to reinforce the representation quality of recommender systems, based on the hypergraph self-discrimination. Extensive experiments on three benchmark datasets demonstrate the superiority of our model over various state-of-the-art recommendation methods, and the robustness against sparse user interaction data. The implementation codes are available at https://github.com/akaxlh/HCCF.

A well-informed recommendation framework could not only help users identify their interested items, but also benefit the revenue of various online platforms (e.g., e-commerce, social media). Traditional recommendation models usually assume that only a single type of interaction exists between user and item, and fail to model the multiplex user-item relationships from multi-typed user behavior data, such as page view, add-to-favourite and purchase. While some recent studies propose to capture the dependencies across different types of behaviors, two important challenges have been less explored: i) Dealing with the sparse supervision signal under target behaviors (e.g., purchase). ii) Capturing the personalized multi-behavior patterns with customized dependency modeling. To tackle the above challenges, we devise a new model CML, Contrastive Meta Learning (CML), to maintain dedicated cross-type behavior dependency for different users. In particular, we propose a multi-behavior contrastive learning framework to distill transferable knowledge across different types of behaviors via the constructed contrastive loss. In addition, to capture the diverse multi-behavior patterns, we design a contrastive meta network to encode the customized behavior heterogeneity for different users. Extensive experiments on three real-world datasets indicate that our method consistently outperforms various state-of-the-art recommendation methods. Our empirical studies further suggest that the contrastive meta learning paradigm offers great potential for capturing the behavior multiplicity in recommendation. We release our model implementation at: https://github.com/weiwei1206/CML.git.

Watching online videos has become more and more popular and users tend to watch videos based on their personal tastes and preferences. Providing a customized ranking list to maximize the user's satisfaction has become increasingly important for online video platforms. Existing personalized search methods (PSMs) train their models with user feedback information (e.g. clicks). However, we identified that such feedback signals may indicate attractiveness but not necessarily indicate relevance in video search. Besides, the click data and user historical information are usually too sparse to train a good PSM, which is different from the conventional Web search containing users' rich historical information. To address these concerns, in this paper we propose a multi-task graph neural network architecture for personalized video search (MGNN-PVS) that can jointly model user's click behaviour and the relevance between queries and videos. To relieve the sparsity problem and learn better representation for users, queries and videos, we develop an efficient and novel GNN architecture based on neighborhood sampling and hierarchical aggregation strategy by leveraging their different hops of neighbors in the user-query and query-document click graph. Extensive experiments on a major commercial video search engine show that our model significantly outperforms state-of-the-art PSMs, which illustrates the effectiveness of our proposed framework.

In this article, we propose a Latent Dirichlet Allocation– (LDA) based topic-graph probabilistic personalization model for Web search. This model represents a user graph in a latent topic graph and simultaneously estimates the probabilities that the user is interested in the topics, as well as the probabilities that the user is not interested in the topics. For a given query issued by the user, the webpages that have higher relevancy to the interested topics are promoted, and the webpages more relevant to the non-interesting topics are penalized. In particular, we simulate a user’s search intent by building two profiles: A positive user profile for the probabilities of the user is interested in the topics and a corresponding negative user profile for the probabilities of being not interested in the the topics. The profiles are estimated based on the user’s search logs. A clicked webpage is assumed to include interesting topics. A skipped (viewed but not clicked) webpage is assumed to cover some non-interesting topics to the user. Such estimations are performed in the latent topic space generated by LDA. Moreover, a new approach is proposed to estimate the correlation between a given query and the user’s search history so as to determine how much personalization should be considered for the query. We compare our proposed models with several strong baselines including state-of-the-art personalization approaches. Experiments conducted on a large-scale real user search log collection illustrate the effectiveness of the proposed models.

Fraud activities in e-commerce, such as spam reviews and fake shopping behaviors, significantly mislead customers' decision making, damage the platforms' reputation, and reduce enterprises' revenue. In recent years, GNN-based models have been widely adopted in fraud detection tasks, which have shown better performance compared to conventional rule-based methods and feature-based models. Most GNN-based models focus on homogeneous graphs, usually including user-to-user, or item-to-item connections. These types of graphs have limitations of eliminating certain types of connections, such as user-item connections. In addition, GNN-based models aggregate neighborhood information based on the assumption that neighbors share the similar structure and content. However, in fraud detection tasks, two major inconsistency issues arise: Severe mixture of structure-inconsistency due to extremely unbalanced positive and negative samples; and mixture of content-inconsistency due to the difference between various item categories. To address the above issues, we propose a Community-based Framework with ATtention mechanism for large-scale Heterogeneous graphs (C-FATH). In order to utilize the entire heterogeneous graph, we directly model on the heterogeneous graph and combine it with homogeneous graphs. The structure-inconsistent nodes are filtered by introducing the community information when constructing neighbors. Content-inconsistent nodes are selected with lower probability by a similarity-based sampling strategy. Further, the model is trained in a multi-task manner that each node type (e.g. user, item, device, order, and review) is associated with a specific loss function. Comprehensive experiments are conducted on two public review datasets and two large-scale datasets from JD.com, and the experimental results demonstrate the effectiveness and scalability of the proposed C-FATH compared to the state-of-the-art approaches.

Purchase intent forecasting, which aims to model user consumption behavior over different categories of items, plays a key role in many services, like online retailing systems, computational advertising and personalized recommendations. While the recently emerged deep neural network models (e.g., recurrent neural network, or attention mechanism) have been proposed to understand user's sequential behavior, we argue that the successes of these methods is largely rely on the data sufficiency. However, the practical purchase forecasting scenarios involve highly sparse data distributions across categories and time. In such cases, one has to deal with the data imbalance problem in order to encode the complex patterns of user purchase behaviors. To tackle this challenge, we develop a Convolutional Hierarchical TRansformer networks (CHTR), to enable the purchase pattern modeling with the multi-grained temporal dynamics, so as to alleviate the data imbalance issue. In our CHTR framework, we develop a multi-grained hierarchical transformer network, to make the learned behavior embeddings be reflective of the multi-level relational structures. Then, a dependency modeling component is proposed to aggregate the multi-relational context signals and capture the underlying dependent structures. Our experiments on real-world datasets show the significant improvements obtained by CHTR over different types of alternative methods.

In feeds recommendation, users are able to constantly browse items generated by never-ending feeds using mobile phones. The implicit feedback from users is an important resource for learning to rank, however, building ranking functions from such observed data is recognized to be biased. The presentation of the items will influence the user's judgements and therefore introduces biases. Most previous works in the unbiased learning to rank literature focus on position bias (i.e., an item ranked higher has more chances of being examined and interacted with). By analyzing user behaviors in product feeds recommendation, in this paper, we identify and introduce context bias, which refers to the probability that a user interacting with an item is biased by its surroundings, to unbiased learning to rank. We propose an Unbiased Learning to Rank with Combinational Propensity (ULTR-CP) framework to remove the inherent biases jointly caused by multiple factors. Under this framework, a context-aware position bias model is instantiated to estimate the unified bias considering both position and context biases. In addition to evaluating propensity score estimation approaches by the ranking metrics, we also discuss the evaluation of the propensities directly by checking their balancing properties. Extensive experiments performed on a real e-commerce data set collected from JD.com verify the effectiveness of context bias and illustrate the superiority of ULTR-CP against the state-of-the-art methods.

E-commerce sites usually leverage taxonomies for better organizing products. The fine-grained categories, regarding the leaf categories in taxonomies, are defined by the most descriptive and specific words of products. Fine-grained product categorization remains challenging, due to blurred concepts of fine grained categories (i.e. multiple equivalent or synonymous categories), instable category vocabulary (i.e. the emerging new products and the evolving language habits), and lack of labelled data. To address these issues, we proposes a novel Neural Product Categorization model---NPC to identify fine-grained categories from the product content. NPC is equipped with a character-level convolutional embedding layer to learn the compositional word representations, and a spiral residual layer to extract the word context annotations capturing complex long range dependencies and structural information. To perform categorization beyond predefined categories, NPC categorizes a product by jointly recognizing categories from the product content and predicting categories from predefined category vocabularies. Furthermore, to avoid extensive human labors, NPC is able to adapt to weak labels, generated by mining the search logs, where the customers' behaviors naturally connect products with categories. Extensive experiments performed on a real e-commerce platform datasets illustrate the effectiveness of the proposed models.

Query intent understanding is a fundamental and essential task in searching, which promotes personalized retrieval results and users' satisfaction. In E-commerce, query understanding is particularly referring to bridging the gap between query representations and product representations. In this paper, we aim to map the queries into the predefined tens of thousands of fine-grained categories extracted from the product descriptions. The problem is very challenging in several aspects. First, a query may be related to multiple categories and to identify all the best matching categories could eventually drive the search engine for high recall and diversity. Second, the same query may have dynamic intents under various scenarios and there is a need to distinguish the differences to promote accurate categories of products. Third, the tail queries are particularly difficult for understanding due to noise and lack of customer feedback information. To better understand the queries, we firstly conduct analysis on the search queries and behaviors in the E-commerce domain and identified the uniqueness of our problem (e.g. longer sessions). Then we propose a Dynamic Product-aware Hierarchical Attention (DPHA) framework to capture the explicit and implied meanings of a query given its context information in the session. Specifically, DPHA automatically learns the bidirectional query-level and self-attentional session-level representations which can capture both complex long range dependencies and structural information. Extensive experimental results on a real E-commerce query data set demonstrate the effectiveness of the proposed DPHA compared to the state-of-art baselines.

Online purchase forecasting is of great importance in e-commerce platforms, which is the basis of how to present personalized interesting product lists to individual customers. However, predicting online purchases is not trivial as it is influenced by many factors including: (i) the complex temporal pattern with hierarchical inter-correlations; (ii) arbitrary category dependencies. To address these factors, we develop a Graph Multi-Scale Pyramid Networks (GMP) framework to fully exploit users' latent behavioral patterns with both multi-scale temporal dynamics and arbitrary inter-dependencies among product categories. In GMP, we first design a multi-scale pyramid modulation network architecture which seamlessly preserves the underlying hierarchical temporal factors--governing users' purchase behaviors. Then, we employ convolution recurrent neural network to encode the categorical temporal pattern at each scale. After that, we develop a resolution-wise recalibration gating mechanism to automatically re-weight the importance of each scale-view representations. Finally, a context-graph neural network module is proposed to adaptively uncover complex dependencies among category-specific purchases. Extensive experiments on real-world e-commerce datasets demonstrate the superior performance of our method over state-of-the-art baselines across various settings.

Citywide abnormal events, such as crimes and accidents, may result in loss of lives or properties if not handled efficiently. It is important for a wide spectrum of applications, ranging from public order maintaining, disaster control and people's activity modeling, if abnormal events can be automatically predicted before they occur. However, forecasting different categories of citywide abnormal events is very challenging as it is affected by many complex factors from different views: (i) dynamic intra-region temporal correlation; (ii) complex inter-region spatial correlations; (iii) latent cross-categorical correlations. In this paper, we develop a Multi-View and Multi-Modal Spatial-Temporal learning (MiST) framework to address the above challenges by promoting the collaboration of different views (spatial, temporal and semantic) and map the multi-modal units into the same latent space. Specifically, MiST can preserve the underlying structural information of multi-view abnormal event data and automatically learn the importance of view-specific representations, with the integration of a multi-modal pattern fusion module and a hierarchical recurrent framework. Extensive experiments on three real-world datasets, i.e., crime data and urban anomaly data, demonstrate the superior performance of our MiST method over the state-of-the-art baselines across various settings.

In probabilistic BM25, term frequency normalization is one of the key components. It is often controlled by parameters $k_1$ and b , which need to be optimized for each given data set. In this paper, we assume and show empirically that term frequency normalization should be specific with query length in order to optimize retrieval performance. Following this intuition, we first propose a new term frequency normalization with query length for probabilistic information retrieval, namely \textttBM25\tiny QL . Then \textttBM25\tiny QL is incorporated into the state-of-the-art models CRTER riptsize 2 and LDA-BM25, denoted as $\textttCRTER riptsize 2 ^\texttt\tiny QL $ and \textttLDA-BM25\tiny QL respectively. A series of experiments show that our proposed approaches \textttBM25\tiny QL , $\textttCRTER riptsize 2 ^\texttt\tiny QL $ and \textttLDA-BM25\tiny QL are comparable to BM25, CRTER riptsize 2 and LDA-BM25 with the optimal b setting in terms of MAP on all the data sets.

Traditional information retrieval (IR) models, in which a document is normally represented as a bag of words and their frequencies, capture the term-level and document-level information. Topic models, on the other hand, discover semantic topic-based information among words. In this paper, we consider term-based information and semantic information as two features of query terms and propose a simple enhancement for ad-hoc IR via topic modeling. In particular, three topic-based hybrid models, LDA-BM25, LDA-MATF and LDA-LM, are proposed. A series of experiments on eight standard datasets show that our proposed models can always outperform significantly the corresponding strong baselines over all datasets in terms of MAP and most of datasets in terms of [email protected] and [email protected] A direct comparison on eight standard datasets also indicates our proposed models are at least comparable to the state-of-the-art approaches.

Traditional pseudo relevance feedback (PRF) models choose top k feedback documents for query expansion and treat those documents equally. When k is determined, feedback terms are selected without considering the reliability of these documents for relevance. Because the performance of PRF is sensitive to the selection of feedback terms, noisy terms imported from these irrelevant documents or partially relevant documents will harm the final results extensively. Intuitively, terms in these documents should be considered less important for feedback term selection. Nonetheless, how to measure the reliability of feedback documents is a difficult problem. Recently, topic modeling has become more and more popular in the information retrieval (IR) area. In order to identify how reliable a feedback document is to be relevant, we attempt to adapt the topical information into PRF. However, topics are hard to be quantified and therefore the identification of topic is usually fuzzy. It is very challenging for integrating the obtained topical information effectively into IR and other text-processing-related areas. Current research work mainly focuses on mining relevant information from particular topics. This is extremely difficult when the boundaries of different topics are hard to define. In this article, we investigate a key factor of this problem, the topic number for topic modeling and how it makes topics “fuzzy.” To effectively and efficiently apply topical information, we propose a new probabilistic framework, “TopPRF,” and three models, TS-COS, TS-EU, and TS-Entropy, via integrating “Topic Space” (TS) information into pseudo relevance feedback. These methods discover how reliable a document is to be relevant through both term and topical information. When selecting feedback terms, candidate terms in more reliable feedback documents should obtain extra weights. Experimental results on various public collections justify that our proposed methods can significantly reduce the influence of “fuzzy topics” and obtain stable, good results over the strong baseline models. Our proposed probabilistic framework, TopPRF, and three topic-space-based models are capable of searching documents beyond traditional term matching only and provide a promising avenue for constructing better topic-space-based IR systems. Moreover, in-depth discussions and conclusions are made to help other researchers apply topical information effectively.

We propose to enhance proximity-based probabilistic retrieval models with more contextual information. A term pair with higher contextual relevance of term proximity is assigned a higher weight. Several measures are proposed to estimate the contextual relevance of term proximity. We assume the top ranked documents from a basic weighting model are more relevant to the query, and calculate the contextual relevance of term proximity using the top ranked documents. We propose a context-sensitive proximity model, and the experimental results on standard TREC data sets show the effectiveness of our proposed model.

Traditionally, in many probabilistic retrieval models, query terms are assumed to be independent. Although such models can achieve reasonably good performance, associations can exist among terms from a human being’s point of view. There are some recent studies that investigate how to model term associations/dependencies by proximity measures. However, the modeling of term associations theoretically under the probabilistic retrieval framework is still largely unexplored. In this article, we introduce a new concept cross term , to model term proximity, with the aim of boosting retrieval performance. With cross terms, the association of multiple query terms can be modeled in the same way as a simple unigram term. In particular, an occurrence of a query term is assumed to have an impact on its neighboring text. The degree of the query-term impact gradually weakens with increasing distance from the place of occurrence. We use shape functions to characterize such impacts. Based on this assumption, we first propose a bigram CRoss TErm Retrieval ( CRTER 2 ) model as the basis model, and then recursively propose a generalized n-gram CRoss TErm Retrieval ( CRTER n ) model for n query terms, where n > 2. Specifically, a bigram cross term occurs when the corresponding query terms appear close to each other, and its impact can be modeled by the intersection of the respective shape functions of the query terms. For an n-gram cross term, we develop several distance metrics with different properties and employ them in the proposed models for ranking. We also show how to extend the language model using the newly proposed cross terms. Extensive experiments on a number of TREC collections demonstrate the effectiveness of our proposed models.

We investigate the effect of rewarding terms according to their locations in documents for probabilistic information retrieval. The intuition behind our approach is that a large amount of authors would summarize their ideas in some particular parts of documents. In this paper, we focus on the beginning part of documents. Several shape functions are defined to simulate the influence of term location information. We propose a Reward Term Retrieval model that combines the reward terms' information with BM25 to enhance probabilistic information retrieval performance.

Term proximity retrieval rewards a document where the matched query terms occur close to each other. Although term proximity is known to be effective in many Information Retrieval (IR) applications, the within-document distribution of each individual query term and how the query terms associate with each other, are not fully considered. In this paper, we introduce a pseudo term, namely Cross Term, to model term proximity for boosting retrieval performance. An occurrence of a query term is assumed to have an impact towards its neighboring text, which gradually weakens with the increase of the distance to the place of occurrence. We use a shape function to characterize such an impact. A Cross Term occurs when two query terms appear close to each other and their impact shape functions have an intersection. We propose a Cross Term Retrieval (CRTER) model that combines the Cross Terms' information with basic probabilistic weighting models to rank the retrieved documents. Extensive experiments on standard TREC collections illustrate the effectiveness of our proposed CRTER model.

In this paper, we present a framework for mining diverging patterns, a new type of contrast patterns whose frequency changes significantly differently in two data sets, e.g., it changes from a relatively low to a relatively high value in one dataset, but from high to low in the other. In this framework, a measure called diverging ratio is defined and used to discover diverging patterns. We use a four-dimensional vector to represent a pattern, and define the pattern's diverging ratio based on the angular difference between its vectors in two datasets. An algorithm is proposed to mine diverging patterns from a pair of datasets, which makes use of a standard frequent pattern mining algorithm to compute vector components efficiently. We demonstrate the effectiveness of our approach on real-world datasets, showing that the method can reveal novel knowledge from large databases.