WWW 2025 Papers

\textit{Federated Learning} (FL) is a distributed machine learning paradigm that allows multiple clients to train models collaboratively without sharing local data. Numerous works have explored security and privacy protection in FL, as well as its integration with blockchain technology. However, existing FL works still face critical issues. \romannumeral1) It is difficult to achieving \textit{poisoning robustness} and \textit{data privacy} while ensuring high \textit{model accuracy}. Malicious clients can launch \textit{poisoning attacks} that degrade the global model. Besides, aggregators can infer private data from the gradients, causing \textit{privacy leakages}. Existing privacy-preserving poisoning defense FL solutions suffer from decreased model accuracy and high computational overhead. \romannumeral2) Blockchain-assisted FL records iterative gradient updates on-chain to prevent model tampering, yet existing schemes are not compatible with practical blockchains and incur high costs for maintaining the gradients on-chain. Besides, incentives are overlooked, where unfair reward distribution hinders the sustainable development of the FL community. In this work, we propose FLock, a robust and privacy-preserving FL scheme based on practical blockchain state channels. First, we propose a lightweight secure \textit{Multi-party Computation} (MPC)-friendly robust aggregation method through quantization, median, and Hamming distance, which could resist poisoning attacks against up to $<50\%$ malicious clients. Besides, we propose communication-efficient Shamir's secret sharing-based MPC protocols to protect data privacy with high model accuracy. Second, we utilize blockchain off-chain state channels to achieve immutable model records and incentive distribution. FLock achieves cost-effective compatibility with practical cryptocurrency platforms, e.g. Ethereum, along with fair incentives, by merging the secure aggregation into a multi-party state channel. In addition, a pipelined \textit{Byzantine Fault-Tolerant} (BFT) consensus is integrated where each aggregator can reconstruct the final aggregated results. Lastly, we implement FLock and the evaluation results demonstrate that FLock enhances robustness and privacy, while maintaining efficiency and high model accuracy. Even with 25 aggregators and 100 clients, FLock can complete one secure aggregation for ResNet in $2$ minutes over a WAN. FLock successfully implements secure aggregation with such a large number of aggregators, thereby enhancing the fault tolerance of the aggregation.

Graph anomaly detection has gained significant attention across various domains, particularly in critical applications like fraud detection in e-commerce platforms and insider threat detection in cybersecurity. Usually, these data are composed of multiple types (e.g., user information and transaction records for financial data), thus exhibiting view heterogeneity. However, in the era of big data, the heterogeneity of views and the lack of label information pose substantial challenges to traditional approaches. Existing unsupervised graph anomaly detection methods often struggle with high-dimensionality issues, rely on strong assumptions about graph structures or fail to handle complex multi-view graphs. To address these challenges, we propose a cluster aware multi-view graph anomaly detection method, called CARE. Our approach captures both local and global node affinities by augmenting the graph's adjacency matrix with the pseudo-label (i.e., soft membership assignments) without any strong assumption about the graph. To mitigate potential biases from the pseudo-label, we introduce a similarity-guided loss. Theoretically, we show that the proposed similarity-guided loss is a variant of contrastive learning loss, and we present how this loss alleviates the bias introduced by pseudo-label with the connection to graph spectral clustering. Experimental results on several datasets demonstrate the effectiveness and efficiency of our proposed framework. Specifically, CARE outperforms the second-best competitors by more than 39% on the Amazon dataset with respect to AUPRC and 18.7% on the YelpChi dataset with respect to AUROC. The code of our method is available at the anonymous GitHub link: https://anonymous.4open.science/r/CARE-demo-1C7F.

With the emergence of edge AI application scenarios such as on-device web search assistants, enhancing the reasoning performance of large language models (LLMs) on edge devices has become an increasingly important topic. Due to the memory and computation limitations of edge devices, edge-cloud collaboration presents a promising solution, which involves deploying smaller LLMs locally while invoking larger-scale LLMs in the cloud. However, how to coordinate these two to balance cost and performance is a challenge. We propose a new collaborative reasoning framework called Division-of-Thoughts (DoT) to fully harness the synergy between locally deployed SLMs and cloud-based LLMs. DOT leverages a Task Decomposer to elicit the inherent planning abilities in language models to decompose user queries into smaller sub-tasks. DoT also employs a Task Scheduler to analyze the pair-wise dependency of sub-tasks and create a dependency graph, facilitating parallel reasoning of sub-tasks and the identification of key steps. To allocate the appropriate model based on the difficulty of sub-tasks, DoT leverages a Plug-and-Play Adapter, which is an additional task head attached to the SLM that does not alter the SLM's parameters. To boost the allocation of the adapter, We also design a self-reinforced tree search algorithm to create a high-qualiy sub-task allocation dataset. Extensive experiments on various benchmarks demonstrate that our DoT significantly reduces LLM costs while maintaining reasoning accuracy. Comparable to the best baseline methods, we reduce the average reasoning time and API costs by 66.12% and 83.57%, respectively. Our code can be accessed via the following link: https://anonymous.4open.science/status/DoT-F17C

Website Fingerprinting (WF) is a traffic analysis technique that aims to identify websites visited by users through the analysis of encrypted traffic patterns. Existing approaches often exhibit limited robustness against network variability and concept drift, resulting in significant performance degradation under real-world HTTPS conditions. Moreover, these methods typically require large-scale training datasets and substantial computational resources, which further increases the complexity of deployment. In this paper, we propose HOLMES, a novel approach that exploits HTTP version parallelism to extract enhanced application-layer features. These features, including the number of web resources transmitting in various HTTP versions, expose up to 4.28 bits of information—surpassing 98\% of previously reported features and demonstrate increased stability across varying network conditions. Complementary to this, we introduce WATSON, a lightweight classification method based on lazy learning, which substantially reduces the dependency on large training datasets. To further enhance the identification accuracy, we incorporate two fingerprint-specific distance metrics that ensure high intra-class similarity. Our experimental evaluation demonstrates that HOLMES \& WATSON significantly enhance both robustness and efficiency, achieving an average accuracy of 87.7\% with only a single sample per website, marking an improvement of over 15\% compared to state-of-the-art methods.

Existing foundation models, such as CLIP, aim to learn a unified embedding space for multimodal data, enabling a wide range of downstream web-based applications like search, recommendation, and content classification. However, these models often overlook the inherent graph structures in multimodal datasets, where entities and their relationships are crucial. For example, in social networks, users are connected through friendships, follows, or interactions, and share content in various modalities like text and images. Multimodal graphs (MMGs) represent such graphs where each node is associated with features from different modalities, while the edges capture the relationships between these entities. On the other hand, existing graph foundation models primarily focus on text-attributed graphs (TAGs) and are not designed to handle the complexities of MMGs. To address these limitations, we propose UniGraph2, a novel cross-domain graph foundation model that enables general representation learning on MMGs, providing a unified embedding space. UniGraph2 employs modality-specific encoders alongside a graph neural network (GNN) to learn a unified low-dimensional embedding space that captures both the multimodal information and the underlying graph structure. We propose a new cross-domain multi-graph pre-training algorithm at scale to ensure effective transfer learning across diverse graph domains and modalities. Additionally, we introduce a new Mixture of Experts (MoE) component to align features from different domains and modalities, ensuring coherent and robust embeddings that unify the information across modalities. Extensive experiments on a variety of multimodal graph tasks demonstrate that UniGraph2 significantly outperforms state-of-the-art models in tasks such as representation learning, transfer learning, and multimodal generative tasks, offering a scalable and flexible solution for learning on MMGs.

Customer-generated reviews on e-commerce websites often contain valuable insights into users' interests in product genres and provide a rich source for mining user preferences. However, most existing neural topic models tend to generate meaningless topics that have low correlations with product genres. Furthermore, they often fail to mine user preferences and discover personalized topic profiles due to the absence of explicit user modeling. To address these limitations, we propose a novel Genre-aware Personalized neural Topic Model (GPTM), which incorporates product genre information into the topic modeling process to ensure the relevance between mined topics and product genres. Moreover, it could produce a personalized topic profile for each user by performing user preference modeling. Extensive experimental results on three publicly available Amazon review corpora validate the effectiveness of the proposed GPTM in genre-aware topic modeling. Furthermore, GPTM surpasses state-of-the-art baselines in user preference mining and generating high-quality personalized topic profiles.

Graph data processing is essential for web-scale applications, including social networks, recommendation systems, and web of things (WoT) systems, where large, sparsely connected graphs dominate. Traditional sparse matrix storage formats like compressed sparse row (CSR) face significant memory and performance bottlenecks in distributed, federated, and edge-based computing environments, which are increasingly central to the web. To address this challenge, we propose GraphCSR, a novel storage format that clusters ver- tices with identical edge degrees and stores only the starting index of each group. This approach minimizes memory overhead and facilitates batch memory access while enhancing overall performance, making it particularly suitable for federated systems and resource-constrained edge nodes. Our experiments across various graph operations and large datasets show that GraphCSR achieves considerable memory savings and performance gains of large-scale, distributed graph processing. When deployed GraphCSR on a production-grade supercomputer with 79,024 computing nodes, it outperforms the top-ranked system on the Graph 500 list, demon- strating its potential for scaling web and WoT graph processing in large-scale distributed computing systems.

Affective polarization, the emotional divide between ideological groups marked by in-group love and out-group hate, has intensified in the United States, driving contentious issues like masking and lockdowns during the COVID-19 pandemic. Despite its societal impact, existing models of opinion change fail to account for emotional dynamics nor offer methods to quantify affective polarization robustly and in real-time. In this paper, we introduce a discrete choice model that captures decision-making within affectively polarized social networks and propose a statistical inference method estimate key parameters---in-group love and out-group hate---from social media data. Through empirical validation from online discussions about the COVID-19 pandemic, we demonstrate that our approach accurately captures real-world polarization dynamics and explains the rapid emergence of a partisan gap in attitudes towards masking and lockdowns. This framework allows for tracking affective polarization across contentious issues has broad implications for fostering constructive online dialogues in digital spaces.

Web APIs are the primary communication form for Web services, with RESTful design being the predominant paradigm. However, RESTful APIs are typically fixed once defined, causing data under- or over-fetching as they can't meet clients' varying Web service needs. While semantic enriched API query languages like GraphQL mitigates this problem, they still face expressiveness limitations for logical operations such as indirect queries and loop traversals. To address this, we propose ORFA (One Request For All), the first in literature that employs WebAssembly (Wasm) as a Web API query language to achieve complete expressiveness of client requests. ORFA's key advantage lies in its use of Wasm's Turing completeness to allow clients to compose arbitrary operations within a single request, thus significantly eliminating redundant data transmission and boosting communication efficiency. Technically, ORFA provides a runtime for executing Wasm query programs and incorporates new module splitting strategies and a caching mechanism customized for integrating Wasm into Web API services, which can enable lightweight code transfer and fast request responses. Experimental results on a realistic testbed and popular Web applications show that ORFA effectively reduces latency by 18.4% and network traffic by 24.5% on average, compared to the state-of-the-art GraphQL.

Web agents have emerged as a promising direction to automate Web task completion based on user instructions, significantly enhancing user experience. Recently, Web agents have evolved from traditional agents to Large Language Models (LLMs)-based Web agents. Despite their success, existing LLM-based Web agents overlook the importance of personalized data (e.g. user profiles and historical Web behaviors) in assisting the understanding of users' personalized instructions and executing customized actions. To overcome the limitation, we first formulate the task of LLM-empowered personalized Web agents, which integrate personalized data and user instructions to personalize instruction comprehension and action execution. To address the absence of a comprehensive evaluation benchmark, we construct a Personalized Web Agent Benchmark (PersonalWAB), featuring user instructions, personalized user data, Web functions, and two evaluation paradigms across three personalized Web tasks. Moreover, we propose a Personalized User Memory-enhanced Alignment (PUMA) framework to adapt LLMs to the personalized Web agent task. PUMA utilizes a memory bank with a task-specific retrieval strategy to filter relevant historical Web behaviors. Based on the behaviors, PUMA then aligns LLMs for personalized action execution through fine-tuning and direct preference optimization. Extensive experiments validate the superiority of PUMA over existing Web agents on PersonalWAB. We release code and data at https://anonymous.4open.science/r/PersonalWAB-CDBF/.

Website Fingerprinting (WF) attacks allow passive eavesdroppers to deduce the websites a user visits by analyzing encrypted traffic, threatening user privacy. While current WF attacks achieve high accuracy, they typically assume single-tab browsing, which is unrealistic as users often open multiple tabs, creating mixed traffic. Existing multi-tab WF approaches require large datasets and frequent retraining due to evolving website content, limiting their practicality. In this paper, we introduce Few-shot Multi-tab Website Fingerprinting (FMWF), a novel approach designed to address the limitations of existing multi-tab WF attacks. FMWF directly tackles the challenges of mixed, overlapping traffic traces generated from multi-tab browsing, leveraging two key innovations: (1) an advanced data augmentation technique that synthesizes realistic multi-tab traffic sequences from easily collected single-tab traces, thereby dramatically reducing the need for large-scale real-world traffic data; and (2) a powerful fine-tuning algorithm based on transfer learning that adapts pre-trained models to new, multi-tab environments with minimal additional data. This two-stage framework enables FMWF to capture the complex effectively, overlapping traffic patterns inherent in multi-tab browsing while maintaining a high level of flexibility and significantly lowering computational and data collection burdens. Our experiments, conducted using real traffic traces collected from three widely-used browsers—Microsoft Edge, Google Chrome, and Tor Browser—highlight the superior performance of FMWF in both closed-world and open-world scenarios. Notably, FMWF achieves a minimum 12.3% improvement in accuracy compared to ARES (SP'23), TMWF (CCS'23), and BAPM (ACSAC'21) in the open-world scenario. The code with related datasets is available at https://anonymous.4open.science/r/FMWF-D164.

Content-based Recommender Systems (CRSs) play a crucial role in shaping user experiences in e-commerce, online advertising, and personalized recommendations. However, due to the large amount of categorical features, the embedding tables used in CRS models pose a significant storage bottleneck for real-world deployment, especially on resource-constrained devices. To address this problem, various embedding pruning methods have been proposed, but most existing ones require expensive retraining steps for each target parameter budget, leading to large computational costs. In reality, this computation cost is a major hurdle in real-world applications with diverse storage requirements, such as federated learning and streaming settings. In this paper, we propose SHApley Value-guided Embedding Reduction (Shaver) as our response. With Shaver, we view the problem from a cooperative game perspective, and quantify each embedding parameter's contribution with Shapley values to facilitate contribution-based parameter pruning. To address the inheriently high computation costs of Shapley values, we propose an efficient and unbiased method to estimate Shapley values of a CRS's embedding parameters. Moreover, in the pruning stage, we put forward a field-aware codebook to mitigate the information loss in the traditional zero-out treatment. Through extensive experiments on three real-world datasets, Shaver has demonstrated competitive performance with lightweight recommendation models across various parameter budgets. The source code is available at https://anonymous.4open.science/r/shaver-E808.

Social network platforms like Reddit are increasingly impacting real-world economics. Meme stocks are a recent phenomena where price movements are driven by retail investors organising themselves via social networks. To study the impact of social networks on meme stocks, the first step is to analyse these networks. Going forward, predicting meme stocks' returns would require to predict dynamic interactions first. This is different from conventional link prediction, frequently applied in e.g. recommendation systems. For this task, it is essential to predict more complex interaction dynamics, such as the exact timing and interaction types like loops. These are crucial for linking the network to meme stock price movements. Dynamic graph embedding (DGE) has recently emerged as a promising approach for modeling dynamic graph-structured data. However, current negative sampling strategies, an important component of DGE, are designed for conventional dynamic link prediction and do not capture the specific patterns present in meme stock-related social networks. This limits the training and evaluation of DGE models in analysing such social networks. To overcome this drawback, we propose novel negative sampling strategies based on the analysis of real meme stock-related social networks and financial knowledge. Our experiments show that the proposed negative sampling strategy can better evaluate and train DGE models targeted at meme stock-related social networks compared to existing baselines.

Thematic analysis (TA) is a widely used qualitative method for identifying underlying meanings within unstructured text. However, TA requires manual processes, which become increasingly labour-intensive and time-consuming as datasets grow. While large language models (LLMs) have been introduced to assist with TA on small-scale datasets, three key limitations hinder their effectiveness on larger datasets. First, current approaches often depend on interactions between an LLM agent and a human coder, a process that becomes challenging with larger datasets. Second, with feedback from the human coder, the LLM tends to mirror the human coder, which provides a narrower viewpoint of the data. Third, existing methods follow a sequential process, where codes are generated for individual samples without recalling or adapting previous codes and associated data, reducing the ability to analyse data holistically. To address these limitations, we propose Thematic-LM, an LLM-based multi-agent system for large-scale computational thematic analysis. Thematic-LM assigns specialised tasks to each agent, such as coding, aggregating codes, and maintaining and updating the codebook. We assign coder agents different identity perspectives to simulate the subjective nature of TA, fostering a more diverse interpretation of the data. We applied Thematic-LM to the Dreaddit dataset and the Reddit climate change dataset to analyse themes related to social media stress and online opinions on climate change. We evaluate the resulting themes based on trustworthiness principles in qualitative research. Our study reveals significant insights, such as assigning different identities to coder agents promotes divergence in codes and themes.

Sarcasm is a subtle expression that indicates the incongruity between literal meanings and factual opinions. For multimodal posts in social medias which consist of both images and texts, sarcasm expressions are even more widespread. Recent works have paid attentions to Multimodal Sarcasm Target Identification (MSTI), which focuses on detecting aspect terms of mockery or ridicule as sarcasm targets. However, the current MSTI benchmark only contains annotations on fine-grained sarcasm targets within sarcastic samples. In practice, it will be featured by two major limitations. First, there lack annotations on non-sarcasm aspects to inform deep models to perceive the semantic difference between sarcasm targets and non-sarcasm aspects. As a result, deep models will tend to incorrectly recognize non-sarcasm aspects as sarcasm targets. Second, there lack non-sarcasm samples to inform deep models to perceive the inherent semantics of sarcasm intentions. Due to the subtle characteristic of sarcasm expressions, models trained with only fine-grained supervision signals cannot thoroughly understand the sarcasm semantics, making the fine-grained task of sarcasm target identification restricted. Motivated by these limitations, this work reconstructs a more comprehensive MSTI benchmark by introducing both fine-grained non-sarcasm aspect annotations for existing sarcasm samples and non-sarcastic samples as non-sarcasm references to enable deep models to clearly perceive the mentioned information during training. Based on the multi-granularity (i.e., both aspect-level and sample-level) non-sarcasm information introduced into this new benchmark, this work further proposes a pluggable Semantics-aware Sarcasm Target Identification mechanism to enhance sarcasm target identification by modeling the overall semantics of sarcasm intentions via an auxiliary sample-level sarcasm recognition task. By modeling the overall semantics of sarcasm intention, deep models can obtain a more comprehensive understanding on sarcasm semantics, leading to improved performance on fine-grained sarcasm target identification. Extensive experiments are conducted to validate our contribution. Both the dataset and implementation code will be released once the paper is accepted.

Code Summarization Model (CSM) has been widely used in code production, such as online and web programming for PHP and Javascript. CSMs are essential tools in code production, enhancing software development efficiency and driving innovation in automated code analysis. However, CSMs face risks of exploitation by unauthorized users, particularly in an online environment where CSMs can be easily shared and disseminated. To address these risks, digital watermarks offer a promising solution by embedding imperceptible signatures within the models to assert copyright ownership and track unauthorized usage. Traditional watermarking for CSM copyright protection faces two main challenges: 1) dataset watermarking methods require separate design of triggers and watermark features based on the characteristics of different programming languages, which not only increases the computation complexity but also leads to a lack of generalization, 2) existing watermarks based on code style transformation are easily identifiable by automated detection, demonstrating poor concealment. To tackle these issues, we propose ModMark, a novel model-level digital watermark embedding method. Specifically, by fine-tuning the tokenizer, ModMark achieves cross-language generalization while reducing the complexity of watermark design. Moreover, we employ code noise injection techniques to effectively prevent trigger detection. Experimental results show that our method can achieve 100% watermark verification rate across various programming languages' CSMs, and the concealment and effectiveness of ModMark can also be guaranteed. Our codes and datasets are available at https://anonymous.4open.science/r/ModMark.

The tension between persuasion and privacy preservation is common in real-world settings. Online platforms should protect the privacy of web users whose data they collect, even as they seek to disclose information about these data (e.g., to advertisers). Similarly, hospitals may share patient data to attract research investments with the obligation to preserve patients' privacy. To address these issues, we study Bayesian persuasion under differential privacy constraints, where the sender must design an optimal signaling scheme for persuasion while guaranteeing the privacy of each agent's private information in the database. To understand how privacy constraints affect information disclosure, we explore two perspectives within Bayesian persuasion: one views the mechanism as releasing a posterior about the private data, while the other views it as sending an action recommendation. The posterior-based formulation leads to privacy-utility tradeoffs, quantifying how the tightness of privacy constraints impacts the sender's optimal utility. For any instance in a common utility function family and a wide range of privacy levels, a significant constant gap in the sender's optimal utility can be found between any two of the three conditions: $\epsilon$-differential privacy constraint, relaxation $(\epsilon,\delta)$-differential privacy constraint, and no privacy constraint. We further geometrically characterize optimal signaling schemes under popular privacy constraints ($\epsilon$-differential privacy, $(\epsilon,\delta)$-differential privacy and Rényi differential privacy), which turns out to be equivalent to finding concave hulls in constrained posterior regions. Finally, we develop polynomial-time algorithms for computing optimal differentially private signaling schemes.

Mobile and Web-of-Things (WoT) devices at the network edge generate vast amounts of data for machine learning applications, yet privacy concerns hinder centralized model training. Federated Learning (FL) allows clients (devices) to collaboratively train a shared model coordinated by a central server without transfer private data, but inherent statistical heterogeneity among clients presents challenges, often leading to a dilemma between clients' needs for personalized local models and the server's goal of building a generalized global model. Existing FL methods typically prioritize either global generalization or local personalization, resulting in a trade-off between these two objectives and limiting the full potential of diverse client data. To address this challenge, we propose a novel framework that simultaneously enhances global generalization and local personalization by Rethinking Information Representation in the Federated learning process (FedRIR). Specifically, we introduce Masked Client-Specific Learning (MCSL), which isolates and extracts fine-grained client-specific features tailored to each client's unique data characteristics, thereby enhancing personalization. Concurrently, the Information Distillation Module (IDM) refines the global shared features by filtering out redundant client-specific information, resulting in a purer and more robust global representation that enhances generalization. By integrating the refined global features with the isolated client-specific features, we construct enriched representations that effectively capture both global patterns and local nuances, thereby improving the performance of downstream tasks on the client. Extensive experiments across diverse datasets demonstrate that FedRIR significantly outperforms state-of-the-art FL methods, achieving up to a 3.93% improvement in accuracy while ensuring robustness and stability in heterogeneous environments.

Personalized news recommendation aims to assist users in finding news articles that align with their interests, which plays a pivotal role in mitigating users’ information overload problem. Despite the breakthrough in personalized news recommendation, the following challenges have been rarely explored: (C1) Comprehending manifold intents coupled within a news article, (C2) Differentiating varying post-read preferences of news articles, and (C3) Addressing the cold-start user problem. To tackle these challenges together, we propose a novel personalized news recommendation framework (CROWN) that employs (1) category-guided intent disentanglement for (C1), (2) consistency-based news representation for (C2), and (3) GNN-enhanced hybrid user representation for (C3). Furthermore, we incorporate a category prediction into the training process of CROWN as an auxiliary task for enhancing intent disentanglement. Extensive experiments on two real-world datasets reveal that (1) CROWN outperforms twelve state-of-the-art news recommendation methods and (2) the proposed strategies significantly improve the accuracy of CROWN.