WWW 2025 Papers

In an edge-assisted federated learning (FL) system, edge servers aggregate the local models from the clients within their coverage areas to produce intermediate models for the production of the global model. This significantly reduces the communication overhead incurred during the FL process. To accelerate model convergence, FedEdge, the state-of-the-art edge-assisted FL system, trains clients' models in local federations when they wait for the global model in each training round. However, our investigation reveals that it drives the global model towards clients with excessive local training, causing model drifts that undermine model performance for other clients. To tackle this problem, this paper presents Maverick, a new edge-assisted FL system that mitigates model drifts by training personalized local models for clients through contrastive local training. It introduces a model-contrastive loss to facilitate personalized local federated training by driving clients' local models away from the global model and close to their corresponding intermediate models. In addition, Maverick includes anomalous models in contrastive local training as negative samples to accelerate the convergence of clients' local models. Extensive experiments are conducted on three widely-used public datasets to comprehensively evaluate the performance of Maverick. Compared to state-of-the-art edge-assisted FL systems, Maverick accelerates model convergence by up to 16.2x and improves model accuracy by up to 12.7%.

Reranking is significant for recommender systems due to its pivotal role in refining recommendation results. To meet diverse reranking requirements in practical applications, numerous reranking models have emerged, which not only prioritize accuracy but also consider additional aspects such as diversity and fairness, etc. However, most of the existing models struggle to strike a harmonious balance between these diverse aspects at the model level. Additionally, the scalability and personalization of these models are often limited by their complexity and a lack of attention to the varying importance of different aspects in diverse reranking scenarios. To address these issues, we propose LLM4Rerank, a comprehensive LLM-based reranking framework designed to bridge the gap between various reranking aspects while ensuring scalability and personalized performance. Specifically, we abstract different aspects into distinct nodes and construct a fully connected graph for LLM to automatically consider aspects like accuracy, diversity, fairness, and more, all in a coherent Chain-of-Thought (CoT) process. To further enhance personalization during reranking, we facilitate a customizable input mechanism that allows fine-tuning of LLM's focus on different aspects according to specific reranking needs. Experimental results on three widely used public datasets demonstrate that LLM4Rerank outperforms existing state-of-the-art reranking models across multiple aspects. The implementation code is available for reproducibility.

Centrality measures are essential for identifying important nodes and edges within a network. In this paper, we focus on two forest-based centrality measures on undirected graphs: forest node centrality (FNC) and forest edge centrality (FEC), which capture the influence of nodes and edges through their participation in spanning forests. Both centrality measures can be represented using entries of the forest matrix. To address the challenge of computing the two measures on large networks, we propose two scalable algorithms from different perspectives. The first algorithm $\textbf{IFGN}$ combines two variance reduction techniques to approximate the entries of the forest matrix, which is applicable to both FNC and FEC. The second algorithm $\textbf{FECE}$ incorporates a new physical interpretation of FEC, allowing for a better overall estimation. We provide error guarantees for both algorithms and demonstrate their efficiency and effectiveness through extensive experiments on various real-world networks.

Multivariate time series anomaly detection is essential for failure management in web application operations, as it directly influences the effectiveness and timeliness of implementing remedial or preventive measures. This task is often framed as a semi-supervised learning problem, where only normal data are available for model training, primarily due to the labor-intensive nature of data labeling and the scarcity of anomalous data. Existing semi-supervised methods often detect anomalies by capturing intra-variate temporal dependencies and/or inter-variate relationships to learn normal patterns, flagging timestamps that deviate from these patterns as anomalies. However, these approaches often fail to capture salient intra-variate temporal and inter-variate dependencies in time series due to their focus on excessively fine granularity, leading to suboptimal performance. In this study, we introduce MtsCID, a novel semi-supervised multivariate time series anomaly detection method. MtsCID employs a dual network architecture: one network operates on the attention maps of multi-scale intra-variate patches for coarse-grained temporal dependency learning, while the other works on variates to capture coarse-grained inter-variate relationships through convolution and interaction with sinusoidal prototypes. This design enhances the ability to capture the patterns from both intra-variate temporal dependencies and inter-variate relationships, resulting in improved performance. Extensive experiments across seven widely used datasets demonstrate that MtsCID achieves performance comparable or superior to state-of-the-art benchmark methods.

Owing to the unprecedented capability in semantic understanding and logical reasoning, the large language models (LLMs) have shown fantastic potential in developing the next-generation sequential recommender systems (RSs). However, on one hand, existing LLM-based sequential RSs mostly separate the index generation from the sequential recommendation, leading to insufficient integration between the semantic information and the collaborative information. On the other hand, the neglect of the user-related information hinders the LLM-based sequential RSs from exploiting the high-order user-item interaction patterns implicating in user behavior. In this paper, we propose the End-to-End Dual Dynamic (ED$^2$) recommender, the first LLM-based sequential recommender system which adopts the dual dynamic index mechanism, targeting at resolving the above limitations simultaneously. The dual dynamic index mechanism can not only assembly the index generation and the sequential recommendation into an unified LLM-backbone pipeline, but also make it practical for the LLM-based sequential recommender to take advantage of the user-related information. Specifically, to facilitate the LLMs comprehension ability to the dual dynamic index, we propose a multi-grained token regulator which constructs alignment supervision based on the LLMs semantic knowledge across multiple representation granularities. Moreover, the associated user collection data and a series of novel instruction tuning tasks are specially customized to exploit the user historical behavior in depth and capture the high-order user-item interaction patterns. Extensive experiments on three public datasets demonstrate the superiority of ED$^2$, achieving an average improvement of 19.41\% in Hit-Rate and 20.84\% in NDCG metric.

The 21st century has already witnessed so many outbreaks with pandemic potential, including SARS (2002), H1N1 (2009), MERS (2012), Ebola (2014), Zika virus (2015), and the COVID-19 pandemic (2019). Using 60 million geotagged Sina Weibo tweets covering over 20 million active accounts, we investigate the collective emotional dynamics on social media in the most recent global pandemic, i.e., COVID-19. This research features two highlights: (1) It focuses on the Chinese population located in the initial epicenter of the pandemic. (2) It examines the initial year after the pandemic outbreak, a critical period where emotions were most intense due to the uncertainty and rapid developments related to the crisis. Using cross-disciplinary methods, we reveal a positive connection between online emotional resonance and geographic proximity, demonstrating a direct mapping between virtual network distances and physical spatial embedding. We propose a percolation-based index to measure the nationwide emotional resonance level with which we illustrate the significant economic impact of the global health issue. Finally, we identify a leader-follower pattern in emotional resonance fluctuations based on time-lag emotion correlations, revealing that less active regions play a crucial role in leading and responding to emotional changes. In the face of long COVID and emerging global health crises, our analysis elucidates how collective emotional resonance evolves, providing potential directions for online opinion interventions during global shocks.

Recent advances in Artificial Intelligence (AI) have introduced a new paradigm in Machine Learning (ML) model development: pre-training of foundation model and domain adaptation. Two groups lead in developing foundation model: closed-source developers and open-source community. As open-source community becomes increasingly engaged, the performance open-source models are catching up with closed-source models. However, this leaves domain deployers into a dilemma: use closed-source models via API access or host open-source models on proprietary hardware. Using closed-source models incurs recurring costs, while hosting open-source models incurs substantial hardware investments and potentially lagging advancements. This paper presents a game-theoretical model to examine the economic incentives behind the deployment choice and the impact of open-source engagement strategy on technology innovation. We find that the deployer consistently opts for closed-source APIs when the open-source community engages in the market reactively by maintaining a fixed performance ratio relative to closed-source advancements. However, open-source models can be favored when a proactive open-source community produces high-performance models independently. Also, we identify conditions under which engagement and competitiveness of the open-source community can foster or inhibit technological progress. These insights offer valuable implications for market regulation and the future of AI model innovation.

With the rise of generative large language models (LLMs) like LLaMA and ChatGPT, these models have significantly transformed daily life and work by providing advanced insights. However, as jailbreak attacks continue to circumvent built-in safety mechanisms, exploiting carefully crafted scenarios or tokens, the safety risks of LLMs have come into focus. While numerous defense strategies—such as prompt detection, modification, and model fine-tuning—have been proposed to counter these attacks, a critical question arises: do these defenses compromise the utility and usability of LLMs for legitimate users? Existing research predominantly focuses on the effectiveness of defense strategies without thoroughly examining their impact on performance, leaving a gap in understanding the trade-offs between LLM safety and performance. Our research addresses this gap by conducting a comprehensive study on the utility degradation, safety elevation, and exaggerated-safety escalation of LLMs with jailbreak defense strategies. We propose _**USEBench**_, a novel benchmark designed to evaluate these aspects, along with _**USEIndex**_, a comprehensive metric for assessing overall model performance. Through experiments on seven state-of-the-art LLMs, we found that mainstream jailbreak defenses fail to ensure both safety and performance simultaneously. Although model-finetuning performs the best overall, their effectiveness varies across LLMs. Furthermore, vertical comparisons reveal that developers commonly prioritize performance over safety when iterating or fine-tuning their LLMs.

Retrieval-Augmented Generation (RAG) has been shown to improve knowledge capabilities and alleviate the hallucination problem of LLMs. The Web is a major source of external knowledge used in RAG systems, and many commercial systems such as ChatGPT and Perplexity have used Web search engines as their major retrieval systems. Typically, such RAG systems retrieve search results, download HTML sources of the results, and then extract plain texts from the HTML sources. Plain text documents or chunks are fed into the LLMs to augment the generation. However, much of the structural and semantic information inherent in HTML, such as headings and table structures, is lost during this plain-text-based RAG process. To alleviate this problem, we propose HtmlRAG, which uses HTML instead of plain text as the format of retrieved knowledge in RAG. We believe HTML is better than plain text in modeling knowledge in external documents, and most LLMs possess robust capacities to understand HTML. However, utilizing HTML presents new challenges. HTML contains additional content such as tags, JavaScript, and CSS specifications, which bring extra input tokens and noise to the RAG system. To address this issue, we propose HTML cleaning, compression, and pruning strategies, to shorten the HTML while minimizing the loss of information. Specifically, we design a two-step block-tree-based pruning method that prunes useless HTML blocks and keeps only the relevant part of the HTML. Experiments on six QA datasets confirm the superiority of using HTML in RAG systems.

As multimedia content continues to grow on the Web, the integration of visual and textual data has become a crucial challenge for Web applications, particularly in recommendation systems. Large Vision Language Models (LVLMs) have demonstrated considerable potential in addressing this challenge across various tasks that require such multimodal integration. However, their application in multimodal sequential recommendation (MSR) has not been extensively studied, despite their potential to significantly enhance the performance of web-based multimodal recommendations. To bridge this gap, we introduce MSRBench, the first comprehensive benchmark designed to systematically evaluate different LVLM integration strategies in web-based recommendation scenarios. We benchmark three state-of-the-art LVLMs, i.e., GPT-4 Vision, GPT4o, and Claude-3-Opus, on the next item prediction task using the constructed Amazon Review Plus dataset, which includes additional item descriptions generated by LVLMs. Our evaluation examines five integration strategies: using LVLMs as recommender, item enhancer, reranker, and various combinations of these roles. The benchmark results reveal that 1) using LVLMs as rerankers is the most effective strategy, significantly outperforming others that rely on LVLMs to directly generate recommendations or only enhance items; 2) GPT-4o consistently achieves the best performance across most scenarios, particularly when employed as a reranker; 3) the computational inefficiency of LVLMs presents a major barrier to their widespread adoption in real-time multimodal recommendation systems. Our codes and datasets will be made publicly available upon acceptance.

Hypergraph neural networks have been promising tools for handling learning tasks involving higher-order data, with notable applications in web graphs, such as modeling multi-way hyperlink structures and complex user interactions. Yet, their generalization abilities in theory are less clear to us. In this paper, we seek to develop margin-based generalization bounds for four representative classes of hypergraph neural networks, including convolutional-based methods (UniGCN), set-based aggregation (AllDeepSets), invariant and equivariant transformations (M-IGN), and tensor-based approaches (T-MPHN). Through the PAC-Bayes framework, our results reveal the manner in which hypergraph structure and spectral norms of the learned weights can affect the generalization bounds, where the key technical challenge lies in developing new perturbation analysis for hypergraph neural networks, which offers a rigorous understanding of how variations in the model's weights and hypergraph structure impact its generalization behavior. Our empirical study examines the relationship between the practical performance and theoretical bounds of the models over synthetic and real-world datasets. One of our primary observations is the strong correlation between the theoretical bounds and empirical loss, with statistically significant consistency in most cases.

Prior studies on Visual Sentiment Understanding (VSU) primarily rely on the explicit scene information (e.g., facial expression) to judge visual sentiments, which largely ignore implicit scene information (e.g., human action, objection relation and visual background), while such information is critical for precisely discovering visual sentiments. Motivated by this, this paper proposes a new Omni-scene driven visual Sentiment Identifying, Locating and Attributing in videos (Omni-SILA) task, aiming to interactively and precisely identify, locate and attribute visual sentiments through both explicit and implicit scene information. Furthermore, this paper believes that this Omni-SILA task faces two key challenges: modeling scene and highlighting implicit scene beyond explicit. To this end, this paper proposes an Implicit-enhanced Causal MoE (ICM) approach for addressing the Omni-SILA task. Specifically, a Scene-Balanced MoE (SBM) and an Implicit-Enhanced Causal (IEC) blocks are tailored to model scene information and highlight the implicit scene information beyond explicit, respectively. Extensive experimental results on our constructed explicit and implicit Omni-SILA datasets demonstrate the great advantage of the proposed ICM approach over advanced Video-LLMs.

Efficient processing of large-scale graphs with billions to trillions of edges is essential for training graph-based large language models (LLMs) in web-scale systems. The increasing complexity and size of these models create significant communication challenges due to the extensive message exchanges required across distributed nodes. Current graph engines struggle to effectively scale across hundreds of computing nodes because they often overlook variations in communication costs within the interconnection hierarchy. To address this challenge, we introduce TuComm, a communication hierarchy-aware engine specifically designed to optimize distributed training of graph-based LLMs. By leveraging hierarchical network topology, TuComm dynamically aggregates and transfers messages, fully accounting for the underlying communication domains, thereby enhancing the efficiency of distributed model training across large-scale systems. We implemented TuComm on top of the message passing interface (MPI), incorporating innovations such as dynamic buffer expansion and active buffer switching to enhance scalability. Evaluations conducted on synthetic and real-world datasets, utilizing up to 79,024 nodes and over 1.2 million processor cores, demonstrate that TuComm surpasses leading graph-parallel systems and state-of-the-art counterparts in both throughput and scalability. Moreover, we have deployed TuComm on a production supercomputer, where it consistently outperforms top solutions on the Graph500 list. These results highlight TuComm’s potential to significantly enhance the efficiency of distributed large-scale graph-based LLM training by optimizing communication among distributed systems, making it an invaluable communication engine for web-scale model training.

Graph-based social recommender systems utilize user-item interaction graphs and user-user social graphs to model user preferences. However, their performance can be limited by redundant and noisy information in these two graphs. Although several recommender studies on data denoising exist, most either rely on heuristic assumptions, which limit their adaptability, or use a single model that combines denoising and recommendation, potentially imposing substantial demands on the model capacity. To address these issues, we propose a dual Graph Denoising Social Recommender (GDSR), which consists of two steps: graph denoising and user preference prediction. \textit{First}, we design a denoising module which exploits a dual denoising model to alleviate noises in the interaction and social graphs by performing multi-step noise removal. We develop three kinds of conditions to guide our dual graph denoising paradigm and propose a cross-domain graph optimization strategy to enhance the structure of denoised graphs. \textit{Second}, we devise a recommender module that employs a dual graph learning structure on denoised graphs to generate recommendations. Moreover, we use additional supervision signals to introduce a graph contrastive learning task, enhancing the recommender module's representation quality and robustness. Experiment results show the effectiveness of our GDSR.

In this paper, we systematically evaluate the effectiveness of existing tools for the dynamic security analysis of client-side JavaScript, focusing in particular on information flow control. Each tool is evaluated in terms of: $(i)$ compatibility, i.e., the ability to process and analyze existing scripts without breaking; $(ii)$ transparency, i.e., the ability to preserve the original script semantics when security enforcement is not necessary; $(iii)$ coverage, i.e., the effectiveness in terms of number of detected information flows; $(iv)$ performance, i.e., the computational overhead introduced by the analysis. Our investigation shows that most of the existing analysis tools are incompatible with the modern Web and the compatibility issues affecting them are not easily fixed. Moreover, transparency issues abound and make us question analysis correctness. This is also confirmed by our coverage evaluation, showing that some tools are unable to detect any information flow on real-world websites, while the remaining tools report significantly different outputs. Finally, we observe that the computational overhead of analysis tools may be significant and can exceed 30x. In the end, out of all the evaluated tools, just one of them (Project Foxhound) is effective enough for practical adoption at scale.

Spreadsheets are ubiquitous across the World Wide Web, playing a critical role in enhancing work efficiency across various domains. Large language model (LLM) has been recently attempted for automatic spreadsheet manipulation but has not yet been investigated in complicated and realistic tasks where reasoning challenges exist (e.g., long horizon manipulation with multi-step reasoning and ambiguous requirements). To bridge the gap with the real-world requirements, we introduce **SheetRM**, a benchmark featuring long-horizon and multi-category tasks with reasoning-dependent manipulation caused by real-life challenges. To mitigate the above challenges, we further propose **SheetAgent**, a novel autonomous agent that utilizes the power of LLMs. SheetAgent consists of three collaborative modules: *Planner*, *Informer*, and *Retriever*, achieving both advanced reasoning and accurate manipulation over spreadsheets without human interaction through iterative task reasoning and reflection. Extensive experiments demonstrate that SheetAgent delivers 20-40\% pass rate improvements on multiple benchmarks over baselines, achieving enhanced precision in spreadsheet manipulation and demonstrating superior table reasoning abilities. More details and visualizations are available at https://sheetagent.github.io. The datasets and source code are available at https://anonymous.4open.science/r/SheetAgent.

In offline reinforcement learning-based recommender systems (RLRS), learning effective state representations is crucial for capturing user preferences that directly impact long-term rewards. However, raw state representations often contain high-dimensional, noisy information and components that are not causally relevant to the reward. Additionally, missing transitions in offline data make it challenging to accurately identify features that are most relevant to user satisfaction. To address these challenges, we propose Policy-Guided Causal Representation (PGCR), a novel two-stage framework for causal feature selection and state representation learning in offline RLRS. In the first stage, we learn a causal feature selection policy that generates modified states by isolating and retaining only the causally relevant components (CRCs) while altering irrelevant components. This policy is guided by a reward function based on the Wasserstein distance, which measures the causal effect of state components on the reward and encourages the preservation of CRCs that directly influence user interests. In the second stage, we train an encoder to learn compact state representations by minimizing the mean squared error (MSE) loss between the latent representations of the original and modified states, ensuring that the representations focus on CRCs and filter out irrelevant variations. We provide a theoretical analysis proving the identifiability of causal effects from interventions, validating the ability of PGCR to isolate critical state components for decision-making. Extensive experiments demonstrate that PGCR significantly improves recommendation performance, confirming its effectiveness for offline RL-based recommender systems.

Graph Neural Networks (GNNs)-based recommendation algorithms typically assume that training and testing data are drawn from independent and identically distributed (IID) spaces. However, this assumption often fails in the presence of out-of-distribution (OOD) data, resulting in significant performance degradation. In this study, we construct a Structural Causal Model (SCM) to analyze interaction data, revealing that environmental confounders (e.g., the COVID-19 pandemic) lead to unstable correlations in GNN-based models, thus impairing their generalization to OOD data. To address this issue, we propose a novel approach, graph representation learning via causal diffusion (CausalDiffRec) for OOD recommendation. This method enhances the model’s generalization on OOD data by eliminating environmental confounding factors and learning invariant graph representations. Specifically, we use backdoor adjustment and variational inference to infer the real environmental distribution, thereby eliminating the impact of environmental confounders. This inferred distribution is then used as prior knowledge to guide the representation learning in the reverse phase of the diffusion process to learn the invariant representa- tion. In addition, we provide a theoretical derivation that proves optimizing the objective function of CausalDiffRec can encourage the model to learn environment-invariant graph representations, thereby achieving excellent generalization performance in recom- mendations under distribution shifts. Our extensive experiments validate the effectiveness of CausalDiffRec in improving the generalization of OOD data, and the average improvement is up to 10.69% on Food, 18.83% on KuaiRec, 22.41% on Yelp2018, and 11.65% on Douban datasets.

The enhance of accuracy in reasoning results of LLMs arouses the community’s interests, wherein pioneering studies investigate post-hoc strategies to rectify potential mistakes. Despite extensive efforts, they are all stuck in a state of resource competition demanding significant time and computing expenses. The cause of the situation lies in failing to identify the fundamental feature of the solutions in this line, coined as the self-denial of LLMs. In other words, LLMs should confidently determine the potential mistakes and carefully execute the targeted correction. As the whole procedure conducts within LLMs, supporting and persuasive references are hard to acquire, while the absence of specific steps towards refining mistakes persists even when errors are acknowledged. In response to the challenges, we present PSSD, which refers to and implements the human psyche structure such that three distinct and interconnected roles contribute to human reasoning. Specifically, PSSD leverages the recent multi-agent paradigm, and is further enhanced with three innovatively conceived roles: (1) the intuition-based id role that provides initial attempts based on benign LLMs; (2) the rule-driven superego role that summarizes rules to regulate the above attempts, and returns specific key points as guidance; and (3) the script-centric ego role that absorbs all procedural information to generate executable script for the final answer prediction. Extensive experiments demonstrate that the proposed design not only better enhance reasoning capabilities, but also seamlessly integrate with current models, leading to superior performance.