A multimodal recommendation system (MRS), which leverages rich multimodal information to model user preferences, has recently attracted significant research interest. Most existing MRSs focus primarily on developing sophisticated encoders for feature extraction, typically relying on simple aggregation of interaction-based features for final predictions. However, this conventional paradigm fails to account for the critical semantic difference between high- and low-rating interactions: while high ratings indicate user preference, low ratings explicitly convey dissatisfaction. Such oversight of negative feedback semantics may significantly limit the system’s recommendation performance. Recently, sign graphs—which model positive and negative feedback signals separately—have gained considerable attention. Inspired by this approach, we propose Sign-Aware Multimodal Graph Recommendation (SiMGR), a novel framework incorporating signed graphs into multimodal recommendation systems. SiMGR fuses multimodal features with signed interactions in a unified graph framework by integrating modality-specific representations and applying user-level thresholds to separate positive and negative subgraphs. A balanced pseudo-edge augmentation strategy is introduced to alleviate sparsity and enhance generalization. Experiments on three public multimodal recommendation datasets show that SiMGR outperforms state-of-the-art baselines, achieving an average 4.28% improvement in NDCG@20.

With the rapid proliferation of short-video platforms and content-driven social networks, sequential recommendation models capable of accurately capturing user interests have become increasingly crucial. Among these, Transformer-based sequential recommendation models have gained widespread adoption due to their superior ability. The positional encoding (PE) in Transformer architectures serves to incorporate positional information into sequences. However, relying solely on original absolute positional information may be insufficient for sequential recommendation models. In contrast, the dwell time after interactions (i.e., the time intervals between consecutive user interactions) provides a more accurate reflection of users' emotional responses and evolving interests. Despite its significance, this aspect has often been overlooked in existing works. To fully utilize this information, our work introduces an adaptive PE method, termed TSAPE (Temporal-Series-Aware Positional Encoding). This approach introduces an innovative modeling of the sequence of time intervals between user interactions, rather than the numerical values of the intervals themselves, thereby capturing real-time feedback on user interests and integrating it with conventional PE mechanisms. Furthermore, we employ multiple layers of one-dimensional convolutional networks and attention mechanisms to endow the features with adaptive capabilities across various time interval scenarios. This enables TSAPE to more accurately capture sequential positional information at any given moment. By enhancing the sequential order information of interactions, TSAPE significantly improves the accuracy of next-item recommendations. We seamlessly integrated our method into several Transformer-based sequential recommendation models and conducted comparisons with state-of-the-art sequential recommendation approaches and widely-used PE methods. The results demonstrate that the integration of TSAPE consistently outperforms the original backbone models and other SOTA methods. The SASRec model integrated with TSAPE achieves an average improvement of 15.61% across three evaluation metrics on four benchmark datasets. Our code has been made publicly available at https://github.com/rongbo-qi/TSAPE_Rec.

Recommendation systems play a central role in modern services, yet often treat item cover images as static attributes, overlooking their influence on user decisions. We introduce the task of cover recommendation and study few-shot, interaction-free selection using multimodal user interest profiles. To address cold-start and sparsity challenges in traditional methods, we propose Multimodal Cover Recommendation (MCRec), a framework that leverages Vision-Language Models (VLMs) for multimodal feature extraction. Our approach includes: (1) a Text-Guided Visual Interest Aggregation network (TGVIA) integrating visual and textual representations; (2) multimodal interest embeddings fused via templated prompts; and (3) a multimodal-driven textual inversion technique enabling training-free generalization to new scenarios. We further propose MCRec+, a fine-tuning variant using hybrid sampling. To support evaluation, we construct three benchmarks and propose two new metrics. Extensive experiments show our methods significantly outperform baselines across datasets, especially with average gains of 3.72% in Recall@1, 1.70% in APMS and 1.25% in MPMS on MCRec. Code and data are publicly available from https://github.com/WeixinZhengRec/MCRec.

The recommendation system, as a widely used and effective tool to alleviate information overload, has been receiving increasing attention regarding its issues of bias and fairness. Many studies have focused on addressing fairness on the item side, targeting item fairness by minimizing exposure discrepancies of items among similar individuals. However, in real-world recommendation scenarios, many relevant items requiring similar exposure to users may exhibit certain dissimilarities while existing methods could not solve the problem. To address this, we define a broader item fairness recommendation issue aimed at improving fairness within specified groups of related individual items, which we term as “intra-group item fairness”. To solve this issue, we propose a Group-oriented Individual Fairness recommendation model called GIFRec. First, we introduce a global exposure balance module to mitigate exposure imbalances at a global level, with the help of multimodal information contained in each item. Then, at the group level, we propose a group fusion embedding representation method, allowing individual items within the same group to adaptively share group information. Additionally, as unfair training opportunities may arise for different items during model training, we propose a general fair intra-group optimization method to reduce individual training biases within the same group. Extensive experiments conducted on four real-world datasets demonstrate the effectiveness of our approach with an average improvement of 11.27% in accuracy and 24.91% in fair ness compared to eight SOTA methods. Our implementation and some details can be found at https://github.com/xderui/GIFRec.

In today’s information-rich era, users rely heavily on recommender systems to identify relevant content. Graph structures, renowned for their ability to model intricate user-content relationships, have become essential to these systems. However, the accuracy of recommendations hinges critically on the quality of node representations within these graphs. Personalized recommendations strive to enhance uniqueness by maximizing the dissimilarity between representations (known as uniformity) while simultaneously ensuring that the representations align closely with the content users engage with (dubbed as alignment). Nevertheless, balancing these conflicting objectives remains a challenge for optimal recommendation performance. To tackle these challenges, we propose an innovative approach called SIURec, which differs significantly from previous studies. Rather than relying on manual weight selection between uniformity and alignment and optimizing uniformity solely on the final representation, SIURec adopts an adaptive adjustment method that learns the optimal weight between uniformity and alignment automatically. By optimizing uniformity at every convolutional layer, SIURec captures users’ sub-interests more effectively, ultimately leading to improved recommendation accuracy. Experimental results on four datasets demonstrate that SIURec achieves superior learning of uniformity (with an average improvement of 4.26% in accuracy compared to eleven SOTA methods) and exhibits robustness across different hyperparameter settings.

Personalized recommendations are integrated into daily life, but providers may want certain items to become more appealing over time through user interactions, yet this issue is often overlooked. The existing works are often based on the assumption that users will passively accept all intermediate sequences or not explore intention modeling in the targeted nudging process. Both of these factors result in suboptimal performance in the proactive recommendation. In this paper, we propose a novel intention-based targeted multi-round proactive recommendation method, dubbed ITMPRec. We first select target items using a pre-match strategy. Then, we employ a multi-round nudging recommendation method, incorporating a module to quantify users' intention-level evolution, helping choose suitable intermediate items. Additionally, we model users' sensitivity to changes caused by these items. Lastly, we propose an LLM agent as a pluggable component to simulate user feedback, offering an alternative to traditional click models by leveraging the agent's external knowledge and reasoning capabilities. Through extensive experiments on four public datasets, we demonstrate the superiority of ITMPRec compared to eight baseline models.

Abstract In recent years, while algorithm-driven recommendation applications have seen widespread use, their negative impacts have also increasingly raised concerns. To gain a more comprehensive understanding of the impact of different recommendation algorithms, we explored the phenomenon of information cocoons, where users are enveloped by homogenized recommended content, in different algorithm-driven recommender systems. We simulated long-term interactions between users and various algorithm-driven recommender systems, trying to recreate multi-stage recommendation scenarios under the influence of complex factors, and explored whether and to what extent users would fall into information cocoons while analyzing the underlying reasons from the perspective of algorithms. We conducted simulation experiments on two real-world recommendation datasets from different fields. The results show that information cocoons is prevalent across different algorithm-driven recommender systems, and the extent of its occurrence varies. Diversity-oriented recommendations can help alleviate information cocoons but are limited in effectiveness. The ability of diversity-aware re-ranking frameworks to alleviate information cocoons is influenced by the basic recommendation models. Not only considering the diversity of the current recommendation list but also the similarity between items and users’ historical consumption content, we proposed a simple and lightweight re-ranking framework called ICMF. Compared to other re-ranking methods, ICMF avoids an average of 12.48% of users encountering homogenized recommended content.

Heterogeneous graph streams represent data interactions in real-world applications and are characterized by dynamic and heterogeneous properties including varying node labels, edge labels and edge weights. The mining of graph streams is critical in fields such as network security , social network analysis , and traffic control. However, the sheer volume and high dynamics of graph streams pose significant challenges for efficient storage and accurate query analysis. To address these challenges, we propose LSketch, a novel sketch technique designed for heterogeneous graph streams. Unlike traditional methods, LSketch effectively preserves the diverse label information inherent in these streams, enhancing the expressive ability of sketches. Furthermore, as graph streams evolve over time, some edges may become outdated and lose their relevance. LSketch incorporates a sliding window model that eliminates expired edges, ensuring that the analysis remains focused on the most current and relevant data automatically. LSketch operates with sub-linear storage space and supports both structure-based and time-sensitive queries with high accuracy. We perform extensive experiments over four real datasets, demonstrating that LSketch outperforms state-of-the-art methods in terms of query accuracy and time efficiency.

Today, mainstream recommendation systems have achieved remarkable success in recommending items that align with user interests. However, limited attention has been paid to the perspective of item providers. Content providers often desire that all their offerings, including unpopular or cold items, aredisplayed and appreciated by users. To tackle the challenges ofunfair exhibition and limited item acceptance coverage, we introduce a novel recommendation perspective that enables items to “select” their most relevant users. We further introduce ItemRec, a straightforward plug-and-play approach that leverages mutual scores calculated by any model. The goal is to maximize the recommendation and acceptance of items by users. Through extensive experiments on three real-world datasets, we demonstrate that ItemRec can enhance valid coverage by up to 38.5% while maintaining comparable or superior recommendation quality. This improvement comes with only a minor increase in model inference time, ranging from 1.5% to 5%. Furthermore, when compared to thirteen state-of-the-art recommendation methods across accuracy, fairness, and diversity, ItemRec exhibits significant advantages as well. Specifically, ItemRec achieves an optimal balance between precision and valid coverage, showcasing an efficiency gain ranging from 1.8 to 45 times compared to other fairness-oriented methodologies.

Abstract Community search (CS) is a vital research area in network science that focuses on discovering personalized communities for query vertices from graphs. However, existing CS methods mainly concentrate on homogeneous or simple attributed graphs, often disregarding complex semantic information and rich contents carried by entities in heterogeneous graphs (HGs). In this paper, we propose a novel problem, namely the “Semantic Network Oriented Community Search with Meta-Structures in Heterogeneous Graphs (SNCS),” which aims to find dense communities that contain the query vertex, with vertices of the same type sharing similar topics. In response to this new problem, we present a novel approach, also named SNCS, representing the first solution employing meta-structures and topic constraints to tackle community search, leveraging both topological and latent features. To overcome the high-time complexity challenge posed by searching through meta-structures, we introduce a unique graph reconstruction technique. Our proposed method’s superiority is validated through extensive evaluations on real-world datasets. The results demonstrate a significant improvement in the quality of the obtained communities, with increases of 3.5–4.4% in clustering coefficient and 5–11% in density while requiring only 4–46% of the running time when compared with the state-of-the-art methods.

Complex networks have been used widely to model a large number of relationships. The outbreak of COVID-19 has had a huge impact on various complex networks in the real world, for example global trade networks, air transport networks, and even social networks, known as racial equality issues caused by the spread of the epidemic. Link prediction plays an important role in complex network analysis in that it can find missing links or predict the links which will arise in the future in the network by analyzing the existing network structures. Therefore, it is extremely important to study the link prediction problem on complex networks. There are a variety of techniques for link prediction based on the topology of the network and the properties of entities. In this work, a new taxonomy is proposed to divide the link prediction methods into five categories and a comprehensive overview of these methods is provided. The network embedding-based methods, especially graph neural network-based methods, which have attracted increasing attention in recent years, have been creatively investigated as well. Moreover, we analyze thirty-six datasets and divide them into seven types of networks according to their topological features shown in real networks and perform comprehensive experiments on these networks. We further analyze the results of experiments in detail, aiming to discover the most suitable approach for each kind of network.

With the rapid development of the information age, many complex systems can be modeled as graphs. However, the unprecedented growth of data makes it extremely difficult for everyday users to process and mine very large graphs, given their limited computing resources such as personal computers and laptops. To address this challenge, we propose selective edge shedding. By estimating the original graph information from the reduced graph, it can accelerate graph algorithms and queries.In this paper, we propose two vertex-degree preserving edge shedding methods, the core of which are to maintain the expected vertex degree, so as to capture the basic characteristics of the network. Both methods allow users to control the size of the reduced graph based on the computing resource constraint. The experimental results show that the methods proposed in this paper can achieve up to 65% higher accuracy on graph analysis tasks compared to the competitive method, while consuming only 26%-57% running time, which fully demonstrates the advantages of the methods proposed in this work.

Nowadays, a graph serves as a fundamental data structure for many applications. As graph edges stream in, users are often only interested in the recent data. In data exploration, how to store and process such massive amounts of graph stream data becomes a significant problem. As vertex and edge attributes are often referred to as labels, we propose a labeled graph sketch that stores real-time graph structural information in sublinear space and supports queries of diverse types. This sketch also supports sliding window queries. We conduct experiments on three real-world datasets, comparing with a state-of-the-art method to show the superiority of our sketch.

Many big data applications today require querying highly dynamic and large-scale data streams for top-k frequent items in the most recent window of any specified size at any time. This is a challenging problem. We show that our novel solution is not only accurate, but it also one to two orders of magnitude faster than previous approaches. Moreover, its memory footprint grows only logarithmically with the window size, rather than linearly as in previous work. Our comprehensive experiments over real-world datasets show that our solution is very effective and scalable. In addition, we devise a concise and efficient solution to a related problem of tracking the frequency of selected items, improving upon previous work by twenty to thirty times in model conciseness while providing the same accuracy and efficiency.

Large-scale information gathering becomes more and more common with the increasing popularity of smartphones, GPS, social networks, and sensor networks. Services based on this real-time data are the logical next step. Service Oriented Stream Systems (SOSS) have a focus on one-time ad hoc queries as op-posed to continuous queries. High availability is crucial in these services. However, data replication has inherent costs, which are particularly burdensome for high rate, often overloaded, SOSS. To provide high availability and to cope with the problem of over-loading the system, we propose a mechanism called soft quorums. Soft quorums incorporate a tuning knob that provides a trade off between query result accuracy and performance. Thus, in essence, soft quorums simultaneously offer high availability and per-query load shedding as needed. This is done in a system-wise optimal way. The parameter choices of soft quorums automatically adapt to dynamic data stream rates and query rates, and minimize the overall system load, given an accuracy requirement. We devise a recovery algorithm and study data quality after recovery. Finally, we conduct a comprehensive experimental study using two real-world and some synthetic datasets.

Sequence data is prevalent, and event processing over sequences is increasingly important in this Big Data era, drawing much attention from both research and industry. In this paper, we address a novel problem, which is to find common event subsequences from two long sequences. This problem is well motivated, with applications in diverse domains. We propose the window-chained longest common subsequence (WCLCS) semantics, and argue that the traditional longest common subsequence (LCS) cannot serve this need. We then devise efficient algorithms to solve this problem by reducing it to a graph problem. We also propose two more methods to improve the performance: one is based on informed search and exploration, and the other is an approximation algorithm with accuracy guarantees. We finally carry out a systematic experimental evaluation using two real-world datasets and some synthetic datasets.

We propose a new local data perturbation method called Aroma. We first show that Aroma is sound in its privacy protection. For that, we devise a realistic privacy game, called the exposure test. We prove that the αβ algorithm, a previously proposed method that is most closely related to Aroma, performs poorly under the exposure test and fails to provide sufficient privacy in practice. Moreover, any data protection method that satisfies ε-differential privacy will succeed in the test. By proving that Aroma satisfies ε-differential privacy, we show that Aroma offers strong privacy protection. We then demonstrate the utility of Aroma by proving that its estimator has significantly smaller errors than the previous state-of-the-art algorithms such as αβ, AM, and FRAPP. We carry out a systematic empirical study using real-world data to evaluate Aroma, which shows its clear advantages over previous methods.

A graph is a fundamental and general data structure underlying all data applications. Many applications today call for the management and query capabilities directly on graphs. Real time graph streams, as seen in road networks, social and communication networks, and web requests, are such applications. Event pattern matching requires the awareness of graph structures, which is different from traditional complex event processing. It also requires a focus on the dynamicity of the graph, time order constraints in patterns, and online query processing, which deviates significantly from previous work on subgraph matching as well. We study the semantics and efficient online algorithms for this important and intriguing problem, and evaluate our approaches with extensive experiments over real world datasets in four different domains.

Although association rule mining has been studied in the literature for quite a while and numerical attributes are prevalent, perhaps surprisingly, the state-of-the-art quantitative association rule mining is rather inefficient and ineffective in discovering all useful rules. In this paper, we propose a novel divide and conquer two-phase algorithm, which is guaranteed to find all good rules efficiently. We further devise an optimization technique for performance. Moreover, we discuss a few issues with managing and using the discovered quantitative association rules. We perform a comprehensive experimental study which shows that our algorithm is one to two orders of magnitude faster than the state-of-the-art one. In addition, we discover significantly more rules that are useful for prediction.

Answering real-time queries, especially over probabilistic data, is becoming increasingly important for service providers. We study anytime query processing algorithms, and extend the traditional query execution plan with a timing component. Our focus is how to determine this timing component, given the queries' deadline constraints. We consider the common multicore processors. Specifically, we propose two query optimization modes: offline periodic optimization and online optimization. We devise efficient algorithms for both offline and online cases followed by a competitive analysis to show the power of our online optimization. Finally, we perform a systematic experimental evaluation using real-world datasets to verify our approaches.

Managing noisy and uncertain data is needed in a great number of modern applications. A major difficulty in managing such data is the sheer number of query result tuples with diverse probabilities. In many cases, users have a preference over the tuples in a deterministic world, determined by a scoring function. Yet it has been a challenging problem to return top-k for uncertain data. Various semantics have been proposed, and they have been shown to give wildly different tuple rankings. In this paper, we propose a completely different approach. Instead of returning users fc tuples, which are merely one point in the complex distribution of top-k tuple vectors, we provide a so-called top-k oracle and users can arbitrarily query it. Intuitively, an oracle is a black box that, whenever given an SQL query, returns its result. Any information we give is based on faithful, best-effort estimates of the ground-truth top-k tuples. This is especially critical in emergency response applications and in monitoring top-k applications. Furthermore, we are the first to provide the nested query capability with the uncertain top-k result being a subquery. We devise various query processing algorithms for top-k oracles, and verify their efficiency and accuracy through a systematic evaluation over real-world and synthetic datasets.