WWW 2025 Papers

Large vision models (LVMs) and large language models (LLMs) are becoming cutting-edge for sequential recommendation, given their success in broad applications. Despite their advantages over traditional approaches, these models suffer more significant performance degradation on tail items against conventional ID-based solutions, which are largely overlooked by recent research. In this paper, we substantiate the above challenges as (1) all-in ground-truth, i.e., the standard cross-entropy (CE) loss focuses solely on the target items while treating all non-ground-truth equally, causing insufficient optimization for tail items, and (2) knowledge transfer tax, i.e., the knowledge encapsulated in LLMs and LVMs dominates the optimization process due to insufficient training for tail items. We propose reweighting and reembedding, a simple yet efficient method to address the above challenges. Specifically, we reinitialize tail item embedding via a Gaussian distribution to alleviate knowledge transfer tax; besides, a reweighting function is incorporated in the CE loss, which adaptively adjusts item weights during training to encourage the model to pay more attention to tail items rather than exclusively optimizing for ground-truth. Overall, our method enables a more nuanced optimization and is mathematically comparable to the direct preference optimization (DPO) in LLMs. Our extensive experiments on three public datasets show our method outperforms fourteen baselines in overall performance and improves the performance on tail items by a large margin. Our code is available at https://anonymous.4open.science/r/R2Rec-0AE0.

As large language models (LLMs) increasingly power web applications, including social networks, the challenge of moderating hate speech has become a critical concern for the Web. These LLM-powered applications, while offering near-human interaction capabilities, are vulnerable to harmful or biased content due to imperfect training data scraped from the Web. Current hate speech detection methods often struggle with limited annotated data, especially for real-time moderation on these platforms. This paper introduces Sentiment-Aided Hate Speech Detection (SAHSD), a novel approach designed to enhance hate speech detection specifically in LLM-powered web applications. By treating hate speech detection as a few-shot learning task, SAHSD utilizes sentiment analysis to refine pre-trained language models (LM) for improved accuracy in recognizing harmful content. SAHSD first employs publicly available sentiment datasets to train a sentiment analysis model, which is then fine-tuned by merging sentiment prompts with hate speech prompts, enabling efficient and accurate detection even with limited training samples. The effectiveness of SAHSD is demonstrated through experiments on widely used web-sourced datasets like SBIC and HateXplain. SAHSD achieves an exceptional F1-score of 0.99 with only 64 training samples and outperforms advanced techniques such as ToKen, MRP, and HARE, with significant improvements of 33% on SBIC and 95% on HateXplain. SAHSD surpasses GPT-4 in generalization performance across multiple datasets, showing an 8% improvement when trained on equal-sized samples. These results underscore SAHSD's potential to enhance content moderation in LLM-driven web platforms, contributing to a safer, more inclusive and accountable Web ecosystem.

In the realm of Intelligent Tutoring System (ITS), the accurate assessment of students' knowledge states through Knowledge Tracing (KT) is crucial for personalized learning. However, due to data bias, $i.e.$, the unbalanced distribution of question groups ($e.g.$, concepts), conventional KT models are plagued by cognitive bias, which tends to result in cognitive underload for overperformers and cognitive overload for underperformers. More seriously, this bias is amplified with the exercise recommendations by ITS. After delving into the causal relations in the KT models, we identify the main cause as the confounder effect of students' historical correct rate distribution over question groups on the student representation and prediction score. Towards this end, we propose a Disentangled Knowledge Tracing (DisKT) model, which separately models students' familiar and unfamiliar abilities based on causal effects and eliminates the impact of the confounder in student representation within the model. Additionally, to shield the contradictory psychology ($e.g.$, guessing and mistaking) in the students’ biased data, DisKT introduces a contradiction attention mechanism. Furthermore, DisKT enhances the interpretability of the model predictions by integrating a variant of Item Response Theory. Experimental results on 11 benchmarks and 3 synthesized datasets with different bias strengths demonstrate that DisKT significantly alleviates cognitive bias and outperforms 14 baselines in evaluation accuracy. Our code and datasets are available at https://anonymous.4open.science/r/DisKT.

Recent years have witnessed the prevalent integration of Large Language Models (LLMs) in various Web applications, such as search engines and recommender systems. As an emerging technique, instruction tuning aims to align pre-trained LLMs as capable chatbots that excel at following human instructions. Previous research indicates that selecting an appropriate subset of a large instruction dataset can enhance the capabilities of LLMs and reduce training costs. However, existing works tend to overlook external correlations between instruction examples during data selection process, which can introduce potential bias and lead to sub-optimal performance. To bridge this gap, we formalize this problem from graph influence maximization perspective and propose Uncertainty-aware influence Maximization (UniMax), a data selection framework that explicitly incorporates the complex inter-dependencies within instruction data. Specifically, we first define a latent instruction graph, treating each instruction example as a graph node and representing their implicit relations as graph edges. Instead of solely relying on heuristic metrics for graph construction, we develop a self-supervised graph learner to uncover the latent structure beyond surface-level feature correlations. After that, we propose an uncertainty-aware influence function to score each example on the instruction graph, allowing a simple greedy algorithm to select a valuable subset that embodies both high influence and uncertainty with an approximation guarantee. Extensive experiments on public datasets show that the proposed approach can significantly enhance model capabilities, underscoring the importance of exploiting data dependencies in instruction data selection.

Privacy-preserving Cross-Domain Recommendation (CDR) has been extensively studied to address the cold-start problem using auxiliary source domains while simultaneously protecting sensitive information. However, existing privacy-preserving CDR methods rely heavily on transferring sensitive user embeddings or behaviour logs, which leads to adopt privacy methods to distort the data patterns before transferring it to the target domain. The distorted information can compromise overall performance during the knowledge transfer process. To overcome these challenges, our approach differs from existing privacy-preserving methods that focus on safeguarding user-sensitive information. Instead, we concentrate on distilling transferable knowledge from insensitive item embeddings, which we refer to as \textbf{prototypes}. Specifically, we propose a conditional model inversion mechanism to accurately distill prototypes for individual users. We have designed a new data format and corresponding learning paradigm for distilling transferable prototypes from traditional recommendation models using model inversion. These prototypes facilitate bridging the domain shift between distinct source and target domains in a privacy-friendly manner. Additionally, they enable the identification of top-k users in the target domain to substitute for cold-start users prediction. We conduct extensive experiments across large real-world datasets, and the results substantiate the effectiveness of PFCDR.

Federated Graph Learning (FedGL) is an emerging Federated Learning (FL) framework that learns the graph data from various clients to train better Graph Neural Networks(GNNs) model. Owing to concerns regarding the security of such framework, numerous studies have attempted to execute backdoor attacks on FedGL, with a particular focus on distributed backdoor attacks. However, all existing methods posting distributed backdoor attack on FedGL only focus on injecting distributed backdoor triggers into the training data of each malicious client, which will cause model performance degradation on original task and is not always effective when confronted with robust federated learning defense algorithms, leading to low success rate of attack. What's more, the backdoor signals introduced by the malicious clients may be smoothed out by other clean signals from the honest clients, which potentially undermining the performance of the attack. To address the above significant shortcomings, we propose a non-intrusive graph distributed backdoor attack(NI-GDBA) that does not require backdoor triggers to be injected in the training data. Our attack trains an adaptive perturbation trigger generator model for each malicious client to learn the natural backdoor from the GNN model downloading from the server with the malicious client's local data. In contrast to traditional distributed backdoor attacks on FedGL via trigger injection in training data, our attack on different datasets such as Molecules and Bioinformatics have higher attack success rate, stronger persistence and stealth, and has no negative impact on the performance of the global GNN model. We also explore the robustness of NI-GDBA under different defense strategies, and based on our extensive experimental studies, we show that our attack method is robust to current federated learning defense methods, thus it is necessary to consider non-intrusive distributed backdoor attacks on FedGL as a novel threat that requires custom defenses. Code is available at an anonymous github repository: https://anonymous.4open.science/r/NI-GDBA-64E5/

A decentralized identity system that can provide users with self-sovereign digital identities to facilitate complete control over their own data is paramount to Web 3.0. The account system on blockchain is an ideal archetype for realizing Web 3.0 decentralized identity. However, a disadvantage of such completely anonymous identity system is that users can create multiple accounts without authentication to obfuscate their activities on the blockchain. In particular, the current anonymous blockchain account system cannot accurately register the social relationships and interactions between real human users, given the amorphous mappings between users and blockchain identities. This work proposes zkBID, a zero-knowledge blockchain-account-based Web 3.0 decentralized identity scheme, to overcome endemic mistrust in blockchain account systems. zkBID links souls (blockchain accounts) to humans (users’ personhood credentials) in a one-to-one manner to truly reflect the social relationships and interactions between humans on the blockchain. zkBID conceals the one-to-one relationships between blockchain accounts and users’ personhood credentials for privacy protection using linkable ring signature. Thus, with zkBID, the users’ blockchain accounts are credible anonymously. Importantly, zkBID is fully decentralized: all user-related data are generated by users and verified by smart contracts on the blockchain. We implemented zkBID and built a blockchain test network for evaluation purposes. Our tests demonstrate the effectiveness of zkBID and suggest proper ways to configure zkBID system parameters.

Deceptive patterns (DPs) are user interface designs deliberately crafted to manipulate users into unintended decisions, often by exploiting cognitive biases for the benefit of companies or services. While numerous studies have explored ways to identify these deceptive patterns, many existing solutions require significant human intervention and struggle to keep pace with the evolving nature of deceptive designs. To address these challenges, we expanded the deceptive pattern taxonomy from security and privacy perspectives, refining its categories and scope. We created a comprehensive dataset of deceptive patterns by integrating existing small-scale datasets with new samples, resulting in 6,725 images and 10,421 DP instances from mobile apps and websites. We then developed DPGuard, a novel automatic tool leveraging commercial multimodal large language models (MLLMs) for deceptive pattern detection. Experimental results show that DPGuard outperforms state-of-the-art methods. Finally, we conducted an extensive empirical evaluation on 2,000 popular mobile apps and websites, revealing that 23.61% of mobile screenshots and 47.27% of website screenshots feature at least one deceptive pattern instance. Through four unexplored case studies that inform security implications, we highlight the critical importance of the unified taxonomy in addressing the growing challenges of Internet deception.

Due to the growing demand for privacy protection, encrypted tunnels have become increasingly popular among mobile app users, which brings new challenges for app fingerprinting (AF)-based network management. Existing methods primarily transfer traditional AF methods to encrypted tunnels directly, ignoring the core obfuscation and re-encapsulation mechanism of encrypted tunnels, thus resulting in unsatisfactory performance. In this paper, we propose DecETT, a dual decouple-based semantic enhancement method for accurate AF under encrypted tunnels. Specifically, DecETT improves AF under encrypted tunnels from two perspectives: app-specific feature enhancement and irrelevant tunnel feature decoupling. Considering the obfuscated app-specific information in encrypted tunnel traffic, DecETT introduces TLS traffic with stronger app-specific information as a semantic anchor to guide and enhance the fingerprint generation for tunnel traffic. Furthermore, to address the app-irrelevant tunnel feature introduced by the re-encapsulation mechanism, DecETT is designed with a dual decouple-based fingerprint enhancement module, which decouples the tunnel feature and app semantic feature from tunnel traffic separately, thereby minimizing the impact of tunnel features on accurate app fingerprint extraction. Evaluation under five prevalent encrypted tunnels indicates that DecETT outperforms state-of-the-art methods in accurate AF under encrypted tunnels, and further demonstrates its superiority under tunnels with more complicated obfuscation. Project page: https://github.com/DecETT/DecETT

Contemporary datasets sourced from the web often adopt a multi-view format, collecting data from diverse sources, domains, or modules. Existing methodologies employed to analyze such datasets frequently overlook or inaccurately allocate the view-weights, pivotal metrics reflecting each view's significance. This work introduces EVA-MVC, a simple yet effective algorithm designed for Equitable View-weight Allocation (EVA) seamlessly integrated with arbitrary Multi-view Clustering (MVC) methods. Within the EVA phase, we establish theoretical connections between view supplementarity and Multi-view Subspace Learning (MSL), leading to the partiton of views into View Communities (VCs) based on these foundational principles. These VCs exhibit internal supplementarity similarities, facilitating Equitable View-weights Allocation through VC-specific MSL. The proposed EVA process precedes and operates independently of traditional or SOTA MVC approaches, requiring no additional processing or specialized design, making it an ideal preprocessing step for MVC applications. Through comprehensive evaluations across diverse multi-view datasets, our findings reveal that our EVA significantly enhances the effectiveness of mainstream MVC frameworks, resulting in a notable performance improvement.

Graph anomaly detection plays a crucial role in identifying nodes that deviate significantly from normal patterns within a graph, with applications spanning various domains such as fraud detection, authorship fraud, and rumor propagation. Traditional methods primarily focus on aggregating information from neighboring nodes and reconstructing the central node based on these aggregated features. The anomaly degree is then calculated by comparing the reconstructed features with the original ones. Despite their effectiveness, these methods face limitations due to the constraints of device performance and the need to protect user privacy. In reality, graph data is often partitioned and distributed across different local clients, which leads to isolated client subgraphs. This partitioning results in incomplete feature aggregation, as the connections between subgraphs are missing, ultimately reducing the performance of anomaly detection models. To overcome these challenges, a federated graph anomaly detection approach based on disentangled representation learning is proposed. This method separates node features into two distinct components: intrinsic features and subgraph style features. By identifying outliers within the subgraph style features, a set of pseudo-nodes is generated and shared across the entire graph. These pseudo-nodes simulate connections between otherwise isolated subgraphs, which enables more comprehensive aggregation of intrinsic features from neighboring nodes. In addition, conditional variational autoencoders (CVAE) are employed alongside contrastive learning strategies to alleviate class imbalance and achieve effective feature disentanglement. These techniques help ensure that anomalous nodes are detected more accurately despite the inherent challenges of federated graph systems. Extensive experiments conducted on six diverse datasets provide compelling evidence of the proposed method's superior performance in federated graph anomaly detection, highlighting its ability to effectively handle incomplete graph structures while maintaining data privacy.

Phishing attacks pose a significant threat to Internet users. Understanding the security posture of phishing infrastructure is crucial for developing effective defense strategies, as it helps identify potential weaknesses that attackers might exploit. Despite extensive research, there may still be a gap in fully understanding these security weaknesses. To address this important issue, this paper presents a longitudinal study of security configurations and vulnerabilities in phishing websites and associated kits. We focus on two main areas: (1) analyzing the security configurations of phishing websites and servers, particularly HTTP headers and application-level security, and (2) examining the prevalence and types of vulnerabilities in phishing kits. We analyze data from 906,731 distinct phishing websites collected over 2.5 years, covering HTML headers, client-side resources, and phishing kits. Our findings suggest that phishing websites often employ weak security configurations, with 88.8% of the 13,344 collected phishing kits containing at least one potential vulnerability, and 12.5% containing backdoor vulnerabilities. These vulnerabilities present an opportunity for defenders to shift from passive defense to active disruption of phishing operations. Our research proposes a new approach to leverage weaknesses in phishing infrastructure, allowing defenders to take proactive actions to disable phishing sites earlier and reduce their effectiveness.

In conversational recommender systems, interactively presenting queries and leveraging user feedback are crucial for efficiently estimating user preferences and improving recommendation quality. Selecting optimal queries in these systems is a significant challenge that has been extensively studied as a sequential decision problem. The expected value of information (EVOI), which computes the expected reward improvement, provides a principled criterion for query selection. However, it is computationally expensive and lacks theoretical performance guarantees. Conversely, conversational bandits offer provable regret upper bounds, but their query selection strategies yield only marginal regret improvements over non-conversational approaches. To address these limitations, we integrate EVOI within the conversational bandit framework by proposing a new conversational mechanism featuring two key techniques: (1) gradient-based EVOI, which replaces the complex Bayesian updates in conventional EVOI with efficient stochastic gradient descent, significantly reducing computational complexity and facilitating theoretical analysis; and (2) smoothed key term contexts, which enhance exploration by adding random perturbations to uncover more specific user preferences. Our approach applies to both Bayesian (Thompson Sampling) and frequentist (UCB) variants of conversational bandits. We introduce two new algorithms, ConTS-EVOI and ConUCB-EVOI, and rigorously prove that they achieve substantially tighter regret bounds, with both algorithms offering a $\sqrt{d}$ improvement in their dependence on the time horizon $T$, where $d$ is the dimension of the feature space. Extensive evaluations on synthetic and real-world datasets validate the effectiveness of our methods.

WebAssembly (Wasm), a binary instruction format designed for efficient cross-platform execution, has rapidly become a foundational web standard, widely adopted in browsers, client-side, and server-side applications. However, its growing popularity has led to an increase in Wasm-targeted malware, including cryptojackers and obfuscated malicious scripts, which pose significant threats to web security. In spite of progress in deep learning based detection methods for Wasm malware, such as MINOS, these approaches face substantial performance degradation in adversarial environments. In our experiments, MINOS’s detection accuracy dropped to 49.90\% under adversarial attacks, revealing critical vulnerabilities. To address this, we introduce \textbf{WasmGuard}, a robust malware detection framework tailored for Wasm. WasmGuard employs FGSM-based adversarial training with prior-based initialization for perturbation bytes in customized sections, coupled with a novel adversarial contrastive learning objective. Using our large-scale dataset, \textbf{WasmMal-15K} (publicly available), WasmGuard outperforms six competing methods, achieving up to 99.20\% Robust Accuracy and 99.93\% Standard Accuracy under PGD-50 adversarial attacks, while maintaining low training overhead. Additionally, we have released \textbf{WebChecker}, a WasmGuard-powered browser plugin, providing real-time protection against malicious Wasm files.

Predicting answers to queries over knowledge graphs is called a complex reasoning task because answering a query requires subdividing it into subqueries. Existing query embedding methods use this decomposition to compute the embedding of a query as the combination of the embedding of the subqueries. This requirement limits the answerable queries to queries having a single free variable and being decomposable, which are called tree-form queries and correspond to the $SROI^-$ description logic. In this paper, we define a more general set of queries, called DAG queries, formulate a description logic corresponding to them, called DAG-DL, propose a query embedding method for them, called DAGE, and a new benchmark to evaluate query embeddings on them. Given the computational graph of a DAG query, DAGE combines the possibly multiple paths between two nodes into a single path with a trainable operator that represents the intersection of relations and learns DAG-DL tautologies. We show that it is possible to implement DAGE on top of existing query embedding methods, and we empirically measure the outstanding improvement of our method over the results of vanilla methods evaluated in tree-form queries that result in relaxing the DAG queries of our proposed benchmark.

IP addresses with web services are crucial in the Internet ecosystem. Classifying these addresses by industry and organization offers valuable insights into the entities utilizing them, enabling more efficient network management and enhanced security. Previous work in website classification and Internet management struggles to offer an IP-level perspective of the industries of web services due to their limited industry categories or potential industry inconsistencies between IP address owners and AS owners. To this end, we present IPdb, an IP-level industry categorization dataset. To construct the dataset, we developed LLMIC, a Large Language Model-based Industry Categorization framework with a precision of nearly 96\%. IPdb serves as a labeled database for future endeavors in developing IP-level industry classifiers, encompassing over 200 million IP addresses. Furthermore, our study indicates that 30\% $\sim$ 50\% of organizations within critical infrastructure industries deploy web servers across multiple ASes. Our study also validates the problem of mismatched granularity in industry categorization at the AS level with 87.83\% ASes in IPv4 and 72.96\% ASes in IPv6 containing IP addresses from different industries.

In web data, advertising images are crucial for capturing user attention and improving advertising effectiveness. Most existing methods generate background for products primarily focus on the aesthetic quality, which may fail to achieve satisfactory online performance. To address this limitation, we explore the use of Multimodal Large Language Models (MLLMs) for generating advertising images by optimizing for Click-Through Rate (CTR) as the primary objective. Firstly, we build targeted pre-training tasks, and leverage a large-scale e-commerce multimodal dataset to equip MLLMs with initial capabilities for advertising image generation tasks. To further improve the CTR of generated images, we propose a novel reward model to fine-tune pre-trained MLLMs through Reinforcement Learning (RL), which can jointly utilize multimodal features and accurately reflect user click preferences. Meanwhile, a product-centric preference optimization strategy is developed to ensure that the generated background content aligns with the product characteristics after fine-tuning, enhancing the overall relevance and effectiveness of the advertising images. Extensive experiments have demonstrated that our method achieves state-of-the-art performance in both online and offline metrics. We will release our code and weights upon acceptance of the paper.

In the digital age, resources such as open-source software and publicly accessible databases form a crucial category of digital public goods, providing extensive benefits across the Internet. However, the inherent non-exclusivity and non-competitiveness of these public goods frequently result in under-provision, a dilemma exacerbated by individuals' tendency to free-ride. This scenario fosters both cooperation and competition among users, leading to the emergence of public goods games. This paper investigates networked public goods games involving heterogeneous players and convex costs to explore solutions of Nash Equilibrium (NE) for this problem. In these games, each player can choose her own effort level, representing the contributions to public goods. We employ network structures to depict the interactions among participants. Each player's utility is composed of a \emph{concave} value component, influenced by collective efforts, and a \emph{convex} cost component, determined solely by individual effort. To the best of our knowledge, this study is the first to explore a networked public goods game with convex costs. Our research begins by examining welfare solutions aimed at maximizing social welfare and ensuring the convergence of pseudo-gradient ascent dynamics. We establish the presence of NE in this model and provide an in-depth analysis of the conditions under which NE is unique. Additionally, we introduce the concept of game equivalence, which expands the range of public goods games that can support a unique NE. We also delve into \emph{comparative statics}, an essential tool in economics, to evaluate how slight modifications in the model—interpreted as monetary redistribution—impact player utilities. In addition, we analyze a particular scenario with a predefined game structure, illustrating the practical relevance of our theoretical insights. Consequently, our research enhances the broader understanding of strategic interactions and structural dynamics in networked public goods games, with significant implications for policy design in internet economic and social networks.

OWL (Web Ontology Language) ontologies, which are able to represent both relational and type facts as standard knowledge graphs and complex domain knowledge in Description Logic (DL) axioms, are widely adopted in domains such as healthcare and bioinformatics. Inspired by the success of knowledge graph embeddings, embedding OWL ontologies has gained significant attention in recent years. Current methods primarily focus on learning embeddings for atomic concepts and roles, enabling the evaluation based on normalized axioms through specially designed score functions. However, they often neglect the embedding of complex concepts, making it difficult to infer with more intricate axioms. This limitation reduces their effectiveness in advanced reasoning tasks, such as Ontology Learning and ontology-mediated Query Answering. In this paper, we propose $\mathcal{EL}^{++}$-closed ontology embeddings which are able to represent any logical expressions in DL $\mathcal{EL}^{++}$ via composition. Furthermore, we develop TransBox, an effective $\mathcal{EL}^{++}$-closed ontology embedding method that can handle many-to-one, one-to-many and many-to-many relations. Our extensive experiments demonstrate that TransBox often achieves state-of-the-art performance across various real-world datasets for predicting complex axioms. Code and data are available at the https://anonymous.4open.science/r/TransBox-F4B7.