Large language model (LLM) agents have emerged as a promising solution for enhancing recommendation systems via user simulation. However, existing studies predominantly resort to prompt-based simulation using frozen LLMs, which frequently results in suboptimal item modeling and user preference learning, thereby ultimately constraining recommendation performance. To address these challenges, we introduce VRAgent-R1, a novel agent-based paradigm that incorporates human-like intelligence in user simulation. Specifically, VRAgent-R1 comprises two distinct agents: the Item Perception (IP) Agent and the User Simulation (US) Agent, designed for interactive user-item modeling. Firstly, the IP Agent emulates human-like progressive thinking based on MLLMs, effectively capturing hidden recommendation semantics in videos. With a more comprehensive multimodal content understanding provided by the IP Agent, the video recommendation system is equipped to provide higher-quality candidate items. Subsequently, the US Agent refines the recommended video sets based on in-depth chain-of-thought (CoT) reasoning and achieves better alignment with real user preferences through reinforcement learning. Experimental results on a large-scale video recommendation benchmark MicroLens-100k have demonstrated the effectiveness of our proposed VRAgent-R1 method, e.g., the IP Agent achieves a 6.0% improvement in NDCG@10, while the US Agent shows approximately 45.0% higher accuracy in user decision simulation compared to state-of-the-art baselines.

Existing video recommendation systems, relying mainly on ID-based embedding mapping and collaborative filtering, often fail to capture in-depth video content semantics. Moreover, most struggle to address biased user behaviors (e.g., accidental clicks, fast skips), leading to inaccurate interest modeling and frequent negative feedback in top recommendations with unclear causes. To tackle this issue, we collect real-world user video-watching sequences, annotate the reasons for users' dislikes, and construct a benchmark dataset for personalized explanations. We then introduce the Agentic Explainable Negative Feedback (ENF) framework, which integrates three core components: (1) the Profile Agent, extracting behavioral cues from users' historical data to derive psychological and personality profiles; (2) the Video Agent, performing comprehensive multimodal video analysis; and (3) the Reason Agent, synthesizing information from the other two agents to predict user engagement and generate explanations. Additionally, we propose the S-GRPO algorithm, enabling the model to progressively address complex tasks during reinforcement fine-tuning. Experimental results on the collected dataset show that our method significantly outperforms state-of-the-art baselines in negative feedback prediction and reason explanation. Notably, it achieves an 8.6% improvement over GPT-4o in reason classification. Deployment on the business platform further validates its benefits: increasing average user watch time by 6.2%, reducing the fast-skip rate by 9.4% , and significantly enhancing user satisfaction.

User feedback is critical for refining recommendation systems, yet explicit feedback (e.g., likes or dislikes) remains scarce in practice. As a more feasible alternative, inferring user preferences from massive implicit feedback has shown great potential (e.g., a user quickly skipping a recommended video usually indicates disinterest). Unfortunately, implicit feedback is often noisy: a user might skip a video due to accidental clicks or other reasons, rather than disliking it. Such noise can easily misjudge user interests, thereby undermining recommendation performance. To address this issue, we propose a novel Group-aware User Behavior Simulation (G-UBS) paradigm, which leverages contextual guidance from relevant user groups, enabling robust and in-depth interpretation of implicit feedback for individual users. Specifically, G-UBS operates via two key agents. First, the User Group Manager (UGM) effectively clusters users to generate group profiles utilizing a ``summarize-cluster-reflect" workflow based on LLMs. Second, the User Feedback Modeler (UFM) employs an innovative group-aware reinforcement learning approach, where each user is guided by the associated group profiles during the reinforcement learning process, allowing UFM to robustly and deeply examine the reasons behind implicit feedback. To assess our G-UBS paradigm, we have constructed a Video Recommendation benchmark with Implicit Feedback (IF-VR). To the best of our knowledge, this is the first multi-modal benchmark for implicit feedback evaluation in video recommendation, encompassing 15k users, 25k videos, and 933k interaction records with implicit feedback. Extensive experiments on IF-VR demonstrate that G-UBS significantly outperforms mainstream LLMs and MLLMs, with a 4.0% higher proportion of videos achieving a play rate > 30% and 14.9% higher reasoning accuracy on IF-VR.

With the advancement of information technology, the Social Internet of Things (SIoT) has fostered the integration of physical devices and social networks, deepening the study of complex interaction patterns. Text attribute graphs (TAGs) capture both topological structures and semantic attributes, enhancing the analysis of complex interactions within the SIoT. However, existing graph learning methods are typically designed for complete attributed graphs, and the common issue of missing attributes in attribute missing graphs (AMGs) increases the difficulty of analysis tasks. To address this, we propose the topology-driven attribute recovery (TDAR) framework, which leverages topological data for AMG learning. TDAR introduces an improved prefilling method for initial attribute recovery using native graph topology. Additionally, it dynamically adjusts propagation weights and incorporates homogeneity strategies within the embedding space to suit AMGs’ unique topological structures, effectively reducing noise during information propagation. Extensive experiments on public datasets demonstrate that TDAR significantly outperforms state-of-the-art methods in attribute reconstruction and downstream tasks, offering a robust solution to the challenges posed by AMGs. The code is available athttps://github.com/limengran98/TDAR.

The task of live video analytics relies on real-time object tracking that typically involves computationally expensive deep neural network (DNN) models. In practice, it has become essential to process video data on edge devices deployed near the cameras. However, these edge devices often have very limited computing resources and thus suffer from poor tracking accuracy. Through a measurement study, we identify three major factors contributing to the performance issue: outdated detection results, tracking error accumulation, and ignorance of new objects. We introduce a novel approach, called Predict & Correct based Tracking, orPCTrack, to systematically address these problems. Our design incorporates three innovative components: 1) a Predictive Detection Propagator that rapidly updates outdated object bounding boxes to match the current frame through a lightweight prediction model; 2) a Frame Difference Corrector that refines the object bounding boxes based on frame difference information; and 3) a New Object Detector that efficiently discovers newly appearing objects during tracking. Experimental results show that our approach achieves remarkable accuracy improvements, ranging from 19.4% to 34.7%, across diverse traffic scenarios, compared to state of the art methods.

Large-scale pre-trained visual-text embedding networks like CLIP have achieved significant progress in cross-modal information retrieval. However, adapting these models for specific scenarios often leads to high training costs and substantial memory demands. In response, we introduce Tiny-CMR, a compact framework designed for cross-modal image-text retrieval. TinyCMR effectively refines the shared semantic space created by pre-trained models, employing a unique blend of reconstruction, modality discrimination, and contrastive learning tasks. By freezing the pre-trained model and integrating a few linear layers, TinyCMR considerably cuts down both training time and space complexity. Experimental results on the MS-COCO and Flickr30K datasets show that TinyCMR achieves accuracy comparable to fine-tuned CLIP and other large-scale models, with over a 400-fold reduction in trainable parameters and a 1000-fold decrease in training time under the same GPU memory constraint. Additionally, TinyCMR proves to be an efficient tool for enhancing the performance of fine-tuned pre-trained models, offering significant improvements with minimal additional training. This positions TinyCMR as a versatile solution for adapting zero-shot models and refining fine-tuned models in cross-modal retrieval tasks.

Cross-domain recommendation is an important method to improve recommender system performance, especially when observations in target domains are sparse. However, most existing techniques focus on single-target or dual-target cross-domain recommendation (CDR) and are hard to be generalized to CDR with multiple target domains. In addition, the negative transfer problem is prevalent in CDR, where the recommendation performance in a target domain may not always be enhanced by knowledge learned from a source domain, especially when the source domain has sparse data. In this study, we propose CAT-ART, a multi-target CDR method that learns to improve recommendations in all participating domains through representation learning and embedding transfer. Our method consists of two parts: a self-supervised Contrastive AuToencoder (CAT) framework to generate global user embeddings based on information from all participating domains, and an Attention-based Representation Transfer (ART) framework which transfers domain-specific user embeddings from other domains to assist with target domain recommendation. CAT-ART boosts the recommendation performance in any target domain through the combined use of the learned global user representation and knowledge transferred from other domains, in addition to the original user embedding in the target domain. We conducted extensive experiments on a collected real-world CDR dataset spanning 5 domains and involving a million users. Experimental results demonstrate the superiority of the proposed method over a range of prior arts. We further conducted ablation studies to verify the effectiveness of the proposed components. Our collected dataset will be open-sourced to facilitate future research in the field of multi-domain recommender systems and user modelling.

Existing benchmark datasets for recommender systems (RS) either are created at a small scale or involve very limited forms of user feedback. RS models evaluated on such datasets often lack practical values for large-scale real-world applications. In this paper, we describe Tenrec, a novel and publicly available data collection for RS that records various user feedback from four different recommendation scenarios. To be specific, Tenrec has the following five characteristics: (1) it is large-scale, containing around 5 million users and 140 million interactions; (2) it has not only positive user feedback, but also true negative feedback (vs. one-class recommendation); (3) it contains overlapped users and items across four different scenarios; (4) it contains various types of user positive feedback, in forms of clicking, liking, sharing, and following, etc; (5) it contains additional features beyond the user IDs and item IDs. We verify Tenrec on ten diverse recommendation tasks by running several classical baseline models per task. Tenrec has the potential to become a useful benchmark dataset for a majority of popular recommendation tasks. Our source codes and datasets will be included in supplementary materials.

Cross-domain recommendation can help alleviate the data sparsity issue in traditional sequential recommender systems. In this paper, we propose the RecGURU algorithm framework to generate a Generalized User Representation (GUR) incorporating user information across domains in sequential recommendation, even when there is minimum or no common users in the two domains. We propose a self-attentive autoencoder to derive latent user representations, and a domain discriminator, which aims to predict the origin domain of a generated latent representation. We propose a novel adversarial learning method to train the two modules to unify user embeddings generated from different domains into a single global GUR for each user. The learned GUR captures the overall preferences and characteristics of a user and thus can be used to augment the behavior data and improve recommendations in any single domain in which the user is involved. Extensive experiments have been conducted on two public cross-domain recommendation datasets as well as a large dataset collected from real-world applications. The results demonstrate that RecGURU boosts performance and outperforms various state-of-the-art sequential recommendation and cross-domain recommendation methods. The collected data will be released to facilitate future research.

We consider sparse representations of signals from redundant dictionaries which are unions of several orthonormal bases. The spark introduced by Donoho and Elad plays an important role in sparse representations. However, numerical computations of sparks are generally combinatorial. For unions of several orthonormal bases, two lower bounds on the spark via the mutual coherence were established in previous work. We constructively prove that both of them are tight. Our main results give positive answers to Gribonval and Nielsen’s open problem on sparse representations in unions of orthonormal bases. Constructive proofs rely on a family of mutually unbiased bases which first appears in quantum information theory.

Similarity join of two datasets P and Q is a primitive operation that is useful in many application domains. The operation involves identifying pairs (p, q), in the Cartesian product of P and Q such that (p, q) satisfies a stipulated similarity condition. In a high-dimensional space, an approximate similarity join based on locality-sensitive hashing (LSH) provides a good solution while reducing the processing cost with a predictable loss of accuracy. A distributed processing framework such as MapReduce allows the handling of large and high-dimensional datasets. However, network cost frequently turns into a bottleneck in a distributed processing environment, thus resulting in a challenge of achieving faster and more efficient similarity join [2]. This paper focuses on collision counting LSH-based similarity join in MapReduce and proposes a network-efficient solution called C2Net to improve the utilization of MapReduce combiners. The solution uses two graph partitioning schemes: (i) minimum spanning tree for organizing LSH buckets replication; and (ii) spectral clustering for runtime collision counting task scheduling. Experiments have shown that, in comparison to the state of the art, the proposed solution is able to achieve 20% data reduction and 50% reduction in shuffle time.

Hardware acceleration through field programmable gate arrays (FPGAs) has recently become a technique of growing interest for many data-intensive applications. Join query is one of the most fundamental database query types useful in relational database management systems. However, the available solutions so far have been beset by higher costs in comparison with other query types. In this paper, we develop a novel solution to accelerate the processing of sort-merge join queries with low match rates. Specifically, our solution makes use of hierarchical indexes to identify result-yielding regions in the solution space in order to take advantage of result sparseness. Further, in addition to one-dimensional equi-join query processing, our solution supports processing of multidimensional similarity join queries. Experimental results show that our solution is superior to the best existing method in a low match rate setting; the method achieves a speedup factor of 4.8 for join queries with a match rate of 5%.

Similarity join of two datasets$P$and$Q$is a primitive operation that is useful in many application domains. The operation involves identifying pairs$(p,q)$, in the Cartesian product of$P$and$Q$such that$(p,q)$satisfies a stipulated similarity condition. In a high-dimensional space, an approximate similarity join based on locality-sensitive hashing (LSH) provides a good solution while reducing the processing cost with a predictable loss of accuracy. A distributed processing framework such as MapReduce allows the handling of large and high-dimensional datasets. However, network cost estimation frequently turns into a bottleneck in a distributed processing environment, thus resulting in a challenge of achieving faster and more efficient similarity join. This paper focuses on collision counting LSH-based similarity join in MapReduce and proposes a network-efficient solution called C2Net to improve the utilization of MapReduce combiners. The solution uses two graph partitioning schemes: (i)minimum spanning treefor organizing LSH buckets replication; and (ii)spectral clusteringfor runtime collision counting task scheduling. Experiments have shown that, in comparison to the state of the art, the proposed solution is able to achieve 20 percent data reduction and 50 percent reduction in shuffle time.

Hardware acceleration through field programmable gate arrays (FPGAs) has recently become a technique of growing interest for many data-intensive applications. Join query is one of the most fundamental database query types useful in relational database management systems. However, the available solutions so far have been beset by higher costs in comparison to other query types. In this paper, we develop a novel solution to accelerate the processing of sort-merge join queries with low match rates. Specifically, our solution makes use of hierarchical indexes to identify result-yielding regions in the solution space in order to take advantage of result sparseness. Further, in addition to one-dimensional equi-join query processing, our solution supports processing of multidimensional similarity join queries. Experimental results show that our solution is superior to the best existing method in a low match rate setting; the method achieves a speedup factor of 4.8 for join queries with a match rate of 5 percent.

Locality sensitive hashing (LSH) is an efficient method for solving the problem of approximate similarity search in high-dimensional spaces. Through LSH, a high-dimensional similarity join can be processed in the same way as hash join, making the cost of joining two large datasets linear. By judicially analyzing the properties of multiple LSH algorithms, we propose a generic method to accelerate the process of joining two large datasets using LSH. The crux of our method lies in the way we identify a set of representative points to reduce the number of LSH lookups. Theoretical analyses show that our proposed method can greatly reduce the number of lookup operations and retain the same result accuracy compared to executing LSH lookups for every query point. Furthermore, we demonstrate the generality of our method by showing that the same principle can be applied to LSH algorithms for three different metrics: the Euclidean distance (QALSH), Jaccard similarity measure (MinHash), and Hamming distance (sequence hashing). Results from experimental studies using real datasets confirm our error analyses and show significant improvements of our method over the state-of-the-art LSH method: to achieve over 0.95 recall, we only need to operate LSH lookups for at most 15% of the query points.

Similarity search on high-dimensional data has been intensively studied for data processing and analytics. Despite its broad applicability, data security and privacy concerns along the trend of data outsourcing have not been fully addressed. In this paper, we investigate privacy-preserving similarity join queries, i.e., a pivotal primitive of similarity search that finds pairwise similar data points across two data sets. We start from locality-sensitive hashing and searchable symmetric encryption, i.e., the most practical techniques for similarity search and encrypted search, respectively. However, the immediate combination of two techniques discloses the distribution of the query set, which is exploitable to compromise the confidentiality of queries. To enhance the security, we propose the frequency hiding query scheme, which allows the server to see the flattened query distribution only. To improve the scalability, we further design the result sharing query scheme, which processes a small portion of query points and shares the results with other nearby points. Besides, we set up a strict constraint to carefully select query points to achieve “as-strong-as-possible” guarantees. We formalize the leakage functions in the context of similarity joins, and conduct rigorous security analysis. We implement and evaluate the proposed query schemes on Azure cloud. Experimental results indicate that they have different tradeoffs on security, efficiency, and accuracy, which can flexibly be used for different deployment scenarios.

Locality sensitive hashing (LSH) is an efficient method for solving the problem of approximate similarity search in highdimensional spaces. Through LSH, a high-dimensional similarity join can be processed in the same way as hash join, making the cost of joining two large datasets linear. By judicially analyzing the properties of multiple LSH algorithms, we propose a generic method to speed up the process of joining two large datasets using LSH. The crux of our method lies in the waywhich we identify a set of representative points to reduce the number of LSH lookups. Theoretical analyzes show that our proposed method can greatly reduce the number of lookup operations and retain the same result accuracy compared to executing LSH lookups for every query point. Furthermore, we demonstrate the generality of our method by showing that the same principle can be applied to LSH algorithms for three different metrics: the Euclidean distance (QALSH), Jaccard similarity measure (MinHash), and Hamming distance (sequence hashing). Results from experimental studies using real datasets confirm our error analyzes and show significant improvements of our method overthe state-of-the-art LSH method: to achieve over 0.95 recall, we only need to operate LSH lookups for at most 15 percent of the query points.

Stylometry is the statistical analyses of variationsin the author's literary style. The technique has been used inmany linguistic analysis applications, such as, author profiling, authorship identification, and authorship verification. Over thepast two decades, authorship identification has been extensivelystudied by researchers in the area of natural language processing. However, these studies are generally limited to (i) a small number of candidate authors, and (ii) documents with similar lengths. In this paper, we propose a novel solution by modeling authorship attribution as a set similarity problem to overcome the two stated limitations. We conducted extensive experimental studies on a real dataset collected from an online book archive, Project Gutenberg. Experimental results show that in comparison to existing stylometry studies, our proposed solution can handlea larger number of documents of different lengths written by alarger pool of candidate authors with a high accuracy.