WWW 2025 Papers

Crowdsourced clustering leverages human input to group items into clusters. The design of tasks for crowdworkers, specifically the number of items presented per query, impacts answer quality and cognitive load. This work investigates the trade-off between query size and answer accuracy, revealing diminishing returns beyond 4-5 items per query. Crucially, we identify contextual bias in crowdworker responses – the likelihood of grouping items depends not only on their similarity but also on the other items present in the query. This structured noise contradicts assumptions made in existing noise models. Our findings underscore the need for more nuanced noise models that account for the complex interplay between items and query context in crowdsourced clustering tasks.

The text-to-image models based on diffusion processes, such as DALL-E, Stable Diffusion, and Midjourney, are capable of transforming texts into detailed images and have widespread applications in art and design. As such, amateur users can easily imitate professional-level paintings by collecting an artist’s work and fine-tuning the model, leading to concerns about artworks’ copyright infringement. To tackle these issues, previous studies either add visually imperceptible perturbation to the artwork to change its underlying styles (perturbation-based methods) or embed posttraining detectable watermarks in the artwork (watermark-based methods). However, when the artwork or the model has been published online, i.e., modification to the original artwork or model retraining is not feasible, these strategies might not be viable. To this end, we propose a novel method for data-use auditing in the text-to-image generation model. The general idea of ArtistAuditor is to identify if a suspicious model has been fine-tuned using specific artists’ artworks by analyzing style-related features. Concretely, ArtistAuditor employs a style extractor to obtain the multi-granularity style representations and treats artworks as samplings of an artist’s style. Then, ArtistAuditor queries a trained discriminator to gain the auditing decisions. The experimental results on six combinations of models and datasets show that ArtistAuditor can achieve high AUC values (> 0.937). By studying ArtistAuditor’s transferability and core modules, we provide valuable insights into the practical implementation. Finally, we demonstrate the effectiveness of ArtistAuditor in real-world cases by an online platform Scenario.1 ArtistAuditor is open-sourced at https://anonymous.4open.science/r/ArtistAuditor.

Spectral graph neural networks are proposed to harness spectral information inherent in graph-structured data through the application of polynomial-defined graph filters, recently achieving notable success in graph-based web applications. Existing studies reveal that various polynomial choices greatly impact spectral GNN performance, underscoring the importance of polynomial selection. However, this selection process remains a critical and unresolved challenge. Although prior work suggests a connection between the approximation capabilities of polynomials and the efficacy of spectral GNNs, there is a lack of theoretical insights into this relationship, rendering polynomial selection a largely heuristic process. To address the issue, this paper examines polynomial selection from an error-sum of function slices perspective. Inspired by the conventional signal decomposition, we represent graph filters as a sum of disjoint function slices. Building on this, we then bridge the polynomial capability and spectral GNN efficacy by proving that the construction error of graph convolution layer is bounded by the sum of polynomial approximation errors on function slices. This result leads us to develop an advanced filter based on trigonometric polynomials, a widely adopted option for approximating narrow signal slices. The proposed filter remains provable parameter efficiency, with a novel Taylor-based parameter decomposition that achieves streamlined, effective implementation. With this foundation, we propose TFGNN, a scalable spectral GNN operating in a decoupled paradigm. We validate the efficacy of TFGNN via benchmark node classification tasks, along with an example graph anomaly detection application to show its practical utility.

Dynamic graph link prediction is widely utilized in the complex web of the real world, such as social networks, citation networks, recommendation systems, etc. Recent Transformer-based link prediction methods on dynamic graphs not only fail to model the fine-grained structures such as triangles with the vanilla Transformers in the graph serialization process, but also amplify the imbalanced distribution of graphs because of their over-estimation of high-degree nodes. To tackle these issues, we propose a Topology-aware Transformer on Dynamic Graph (TopDyG) for link prediction, consisting of a topology injected Transformer (Ti-Transformer) and a mutual information learning (Mi-Learning). % mainly consisting of two components, i.e., the topology-injected (Ti) Transformer and the mutual information (Mi) learning module. The Ti-Transformer explores the explicit structure of serialized graphs, capturing the topological features. The Mi-Learning mines the relationship between nodes by modeling the mutual information with a prior knowledge, alleviating the over-estimation of high-degree nodes when applying the Transformer-based models for the dynamic graph link prediction task. Extensive experiments on four public datasets containing both transductive and inductive settings present the superiority of our proposal. In particular, TopDyG presents an improvement of 43.27% and 28.75% against the state-of-the-art baselines in terms of NDCG and Jaccard, respectively. The advantages are especially obvious on the high-density graphs.

A vast amount of user behavior data is constantly accumulating on today's large recommendation platforms, recording users' various interests and tastes. Preserving knowledge from the old data while new data continually arrives is a vital problem for recommender systems. Existing approaches generally seek to save the knowledge implicitly in the model parameters. However, such a parameter-centric approach lacks scalability and flexibility---the capacity is hard to scale, and the knowledge is inflexible to utilize. Hence, in this work, we propose a framework that turns massive user behavior data to retrievable knowledge (D2K). It is a data-centric approach that is model-agnostic and easy to scale up. Different from only storing unary knowledge such as the user-side or item-side information, D2K propose to store ternary knowledge for recommendation, which is determined by the complete recommendation factors---user, item, and context. The knowledge retrieved by target samples can be directly used to enhance the performance of any recommendation algorithms. Specifically, we introduce a Transformer-based knowledge encoder to transform the old data into knowledge with the user-item-context cross features. A personalized knowledge adaptation unit is devised to effectively exploit the information from the knowledge base by adapting the retrieved knowledge to the target samples. Extensive experiments on two public datasets show that D2K significantly outperforms existing baselines and is compatible with a major collection of recommendation algorithms.

Knowledge Graph (KG) reasoning aims to predict missing entities in incomplete triples, which requires adequate structural information to derive accurate embeddings. However, KGs in the real world are not as dense as the idealized benchmarks, where sparse graph structures restrict the comprehensive structural information for superior performance. Although the logical semantics in KGs shows its potential in alleviating the impact of structural sparsity, there still exist some challenges. The deficient supervision and the semantic gap of logic make it difficult to introduce logical semantics in sparse KG reasoning. To this end, we propose a novel KG reasoning approach LoLLM injecting logic with the supervised information supplied by the Large Language Model (LLM), which is proved to be effective in evaluating and scoring. Firstly, LoLLM derives structural embeddings employing a graph convolutional network (GCN) with relation-aware and triple-aware attention. LoLLM secondly constructs reasoning paths instantiated from the first-order logics extracted from sparse KGs, and injects the logical semantics by a designed LLM-enhanced tuning strategy. We propose a textual loss (TL) and a logical loss (LL) in the optimization and obtain logical tuning embeddings of KG in this process. Finally, LoLLM fuses structural embeddings from the GCN and logical tuning embeddings from the LLM-enhanced tuning for scoring and incomplete triple prediction. Extensive experiments on two sparse KGs and a benchmark show that LoLLM outperforms state-of-the-art structure-based and Language Model (LM)-augmented baselines. Moreover, the logics with corresponding confidences provide explicit explanations as an interpretable paradigm.

Money laundering is the process that intends to legalize the income derived from illicit activities, thus facilitating their entry into the monetary flow of the economy without jeopardizing their source. It is crucial to identify such activities accurately and reliably in order to enforce anti-money laundering (AML). Despite considerable efforts to AML, a large number of such activities still go undetected. Rule-based methods were first widely used in the early days and still be widely used in existing detection systems. With the rise of machine learning, graph-based learning methods have gained prominence in detecting illicit accounts by analyzing money transfer graphs between accounts. However, existing approaches work based on the prerequisite that the transaction graph is centralized, while in practice, money laundering activities usually span multiple financial institutions. Due to regulatory, legal, commercial, and customer privacy concerns, institutions tend not to share data, limiting their utility in practical usage. In this paper, we propose the first algorithm that supports performing AML over multiple institutions while protecting the security and privacy of local data. To evaluate, we construct Alipay-ECB, a real-world dataset comprising digital transactions from Alipay, the world’s largest mobile payment platform, alongside transactions from E-Commerce Bank (ECB). The dataset includes over 200 million accounts and 300 million transactions, covering both intra-institution transactions and those between Alipay and ECB. This makes it the largest real-world transaction graph available for analysis. The experimental results demonstrate that our methods can effectively identify cross-institution money laundering subgroups. Additionally, experiments on synthetic datasets also demonstrate that our method is efficient, requiring only a few minutes on datasets with millions of transactions.

Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy, which is a fundamental web text mining task with broad applications such as web content analysis and semantic indexing. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with a minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) show competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, **T**axonomy **E**nrichment and **L**LM-**E**nhanced weakly-supervised hierarchical text **Class**ification, which combines the general knowledge of LLMs and task-specific features mined from an unlabeled corpus. TELEClass automatically enriches the raw taxonomy with class-indicative features for better label space understanding and utilizes novel LLM-based data annotation and generation methods specifically tailored for the hierarchical setting. Experiments show that TELEClass can significantly outperform previous strong baselines while also achieving comparable performance to zero-shot prompting of LLMs with drastically less inference cost.

Memes serve as a powerful medium of expression in the digital age, shaping cultural discourse and conveying ideas succinctly and engagingly. However, their potential for social abuse highlights the importance of developing effective methods to detect harmful content within memes. Recent studies on memes have focused on transforming images into textual captions using large language models (LLMs). However, these approaches often result in non-informative captions. Furthermore, previous methods have only been tested on limited datasets, providing insufficient evidence of their robustness. To address these limitations, we present a multimodal, agent-based framework designed to generate informative visual descriptions of memes by asking insightful questions to improve visual descriptions in zero-shot visual question-answering settings. Specifically, we leverage an LLM as agents with distinct roles and a large multimodal model (LMM) as a vision expert. These agents first analyze the images and then ask informative questions related to potential social abuse in memes to obtain high-quality answers about the images. Through continuous discussion guided by instructional prompts, the agents gather high-quality information while repeatedly acquiring image data from the LMM, which helps detect social abuse in memes. Finally, the discussion history and basic information are classified using the LLM to obtain the final prediction results in a zero-shot setting. Experimental results on a meme benchmark dataset sourced from 5 diverse meme datasets, comprising 6,626 memes spanning 5 tasks of varying complexity related to social abuse, demonstrate that our framework outperforms state-of-the-art methods, with detailed comparative analysis and ablation studies, further validating its generalizability and ability to retrieve more relevant information for detecting social abuse in memes.

CAPTCHAs have long been essential tools for protecting applications from automated bots. Initially designed as simple questions to distinguish humans from bots, they have become increasingly complex to keep pace with the proliferation of CAPTCHA-cracking techniques employed by malicious actors. However, with the advent of advanced large language models (LLMs), the effectiveness of existing CAPTCHAs is now being undermined. To address this issue, we have conducted an empirical study to evaluate the performance of multimodal LLMs in solving CAPTCHAs and to assess how many attempts human users typically need to pass them. Our findings reveal that while LLMs can solve most CAPTCHAs, they struggle with those requiring complex reasoning—a type of CAPTCHA that also presents significant challenges for human users. Interestingly, our user study showed that the majority of human participants required a second attempt to pass these reasoning CAPTCHAs, a finding not previously reported in existing research. Based on the findings of our empirical study, we introduce IllusionCAPTCHA, an innovative approach designed to be "Human-Easy but AI-Hard". This new CAPTCHA employs visual illusions to create tasks that are intuitive for humans but highly confusing for AI models. Furthermore, we developed a structured, step-by-step method that guides LLMs toward making specific incorrect choices, thereby reducing their ability to bypass CAPTCHA systems successfully. Our evaluation shows that IllusionCAPTCHA can effectively deceive LLMs 100\% of the time. Moreover, our structured design significantly increases the likelihood of AI errors when attempting to solve these challenges. Results from our user study indicate that 86.95\% of participants successfully passed the CAPTCHA on their first attempt, outperforming other CAPTCHA systems.

Projecting knowledge graph queries into an embedding space using geometric models (points, boxes and spheres) can help to answer queries for large incomplete knowledge graphs. In this work, we propose a symbolic learning-free approach using fuzzy logic to address the shape-closure problem that restricted geometric-based embedding models to only a few shapes (e.g. ConE) for answering complex logical queries. The use of symbolic approach facilitates non-closure geometric models (e.g. point, box) to handle logical operators (including negation). This enabled our newly proposed spherical embeddings (SpherE) in this work to use a polar coordinate system to effectively represent hierarchical relation. Results show that the SpherE model can answer existential positive first-order logic and negation queries. We show that SpherE significantly outperforms the point and box embeddings approaches while generating semantically meaningful hierarchy-aware embeddings.

Large language models (LLMs) demonstrate significant potential in various applications; however, they are susceptible to generating hallucinations, which can lead to the spread of misinformation online. Existing studies address hallucination detection by (1) employing reference-based methods that consult external resources for verification or (2) utilizing reference-free methods that mainly estimate answer uncertainty based on LLM's internal states. However, reference-based methods incur significant costs and can be infeasible for obtaining reliable external references. Besides, existing uncertainty estimation (UE) methods often overlook the impact of scenario backgrounds inherited from the query's lexical resources, leading to noise in UE. In almost all real-world applications, users care about the uncertainty concerning semantics or facts instead of the query's scenario information. Therefore, we argue that mitigating scenario-related noise and focusing on semantic information can yield a more desirable UE. In this paper, we introduce a plug-and-play scenario-independent framework to enhance unsupervised UE in LLMs by removing scenario-related noise and focusing on semantic information. This framework is compatible with most existing UE methods, as it leverages only the existing UE methods' outputs. Specifically, we design a scenario-specific sampling to paraphrase queries, maintaining their common semantics while diversifying the scenario distribution. Subsequently, to estimate the contribution of the common semantics, we design a factor analysis (FA) model to disentangle the UE score obtained from the given UE method into a combination of multiple latent factors, which represent the contribution of the common semantics and scenario-related noise. By solving the FA model, we decompose the impact of the most significant factor to approximate the uncertainty caused by the common semantics, thus achieving scenario-independent UE. Extensive experiments and analysis across multiple models and datasets demonstrate the effectiveness of our approach.

The Budget Limited Coreness Maximization (BLCM) problem aims to enhance average user engagement by activating a limited number of connections, i.e., inserting up to b edges to maximize the coreness gain of all vertices in a graph. Due to the cascading feature, we prove the BLCM is NP-hard, APX-hard, and not submodular, meaning greedy sequential edge insertion fails to deliver satisfactory results. As a result, solving BLCM requires combinatorial edge insertion and must face the combinatorial exploration difficulty. This paper proposes the first effective and polynomial-time approach to BLCM. It embeds local combinatorial optimization into global greedy search to boost the benefits of combinatorial optimization while restricting its complexity. Specifically, we propose efficient methods to evaluate the cascaded coreness improvements of two local combinatorial strategies, i.e., when a leader or a group of nodes increase their coreness values via local edge insertion. Note that the key difficulty lies in evaluating the cascading effects. Based on these, we propose three efficient combinatorial edge insertion strategies: (1) Leader-Centric Greedy Insertion (LCGI), (2) Group-Centric Greedy Insertion (GCGI), and (3) a Leader-Group Balance (LGB) insertion. LCGI greedily finds the most influential leader that can produce the highest coreness gain together with its followers. GCGI finds the most influential group that can promote the most coreness gain. LGB combines the two strategies to select edge combinations adaptively. We prove the low complexity of LCGI, GCGI and LGB. Experiments conducted on 13 real-world datasets highlight their practical utility and superiority over existing approaches.

The Automatic Certificate Management Environment (ACME) protocol automates the SSL certificate issuance and renewal process, streamlining large-scale certificate management. Its caching mechanism allows Certificate Authorities (CAs) to store domain validation results for up to 30 days. While this mechanism reduces the burden of re-validation on the CA server, it also introduces a vulnerability where attackers can bypass domain validation using stolen ACME account credentials. In this paper, we introduce the ACME $Authz$ Cache Attack, a method that enables attackers to request fraudulent certificates without domain control. We demonstrate that even Let's Encrypt, the world’s largest CA, is susceptible to this vulnerability. To address this issue, we propose ACME++, an enhanced version of the protocol that binds the ACME account to the client’s IP address and a unique ID, ensuring re-validation for each certificate request from a new client. Our implementation of ACME++ shows that it effectively mitigates this attack with minimal impact on server performance.

Building location embedding from web-sourced satellite imagery has emerged as an enduring research focus in web mining. However, most existing methods are inherently constrained by their reliance on discrete, sparse sampling strategies, failing to capture the essential spatial continuity of geographic spaces. Moreover, the presence of confounding factors in satellite images can distort the perception of actual objects, leading to semantic discontinuity in the embeddings. In this work, we propose **SatCLE**, a novel framework for Continuous Location Embeddings leveraging Satellite imagery. Specifically, to address the out-of-distribution query challenge of spatial continuity, we propose a geospatial refinement strategy comprising stochastic perturbation continuity expansion and graph propagation fusion, which transforms discrete geospatial coordinates into a continuous space. To mitigate the effects of confounders on semantic continuity, we introduce causal refinement, integrating causal theory to localize and eliminate spurious correlations arising from the environmental context. Through extensive experiments, **SatCLE** shows state-of-the-art performance, exhibiting superior spatial coherence and semantic fidelity across diverse geospatial tasks.

Empathetic Response Generation (ERG) is one of the key tasks of the affective computing area, which aims to produce emotionally nuanced and compassionate responses to user's queries. However, existing ERG research is predominantly confined to the singleton text modality, limiting its effectiveness since human emotions are inherently conveyed through multiple modalities. To combat this, we introduce an avatar-based Multimodal ERG (MERG) task, entailing rich text, speech, and facial vision information. We first present a large-scale high-quality benchmark dataset, \textbf{AvaMERG}, which extends traditional text ERG by incorporating authentic human speech audio and dynamic talking-face avatar videos, encompassing a diverse range of avatar profiles and broadly covering various topics of real-world scenarios. Further, we deliberately tailor a system, named \textbf{Empatheia}, for MERG. Built upon a Multimodal Large Language Model (MLLM) with multimodal encoder, speech and avatar generators, Empatheia performs end-to-end MERG, with Chain-of-Empathetic reasoning mechanism integrated for enhanced empathy understanding and reasoning. Finally, we devise a list of empathetic-enhanced tuning strategies, strengthening the capabilities of emotional accuracy and content, avatar-profile consistency across modalities. Experimental results on AvaMERG data demonstrate that Empatheia consistently shows superior performance than baseline methods on both textual ERG and MERG. Overall, this work is expected to pioneer the MERG research by introducing a novel benchmark and an end-to-end model, laying a solid foundation for future advancements in multimodal empathetic response generation.

This paper proposes CrossLink, a novel framework for cross-domain link prediction. CrossLink learns the evolution pattern of a specific downstream graph and subsequently makes pattern-specific link predictions. It employs a technique called \textit{conditioned link generation}, which integrates both evolution and structure modeling to perform evolution-specific link prediction. This conditioned link generation is carried out by a transformer-decoder architecture, enabling efficient parallel training and inference. CrossLink is trained on extensive dynamic graphs across diverse domains, encompassing 6 million dynamic edges. Extensive experiments on eight untrained graphs demonstrate that CrossLink achieves state-of-the-art performance in cross-domain link prediction. Compared to advanced baselines under the same settings, CrossLink shows an average improvement of 11.40% in Average Precision across eight graphs. Impressively, it surpasses the fully supervised performance of 8 advanced baselines on 6 untrained graphs.

As the complexity and frequency of cyberattacks, such as Advanced Persistent Threats (APTs) and ransomware, continue to escalate, traditional anomaly detection methods have proven inadequate in addressing these sophisticated, multi-faceted threats. Recently, Host Provenance Graphs (HPGs) have played a crucial role in analyzing system-level interactions, detecting anomalous behaviors, and tracing attack chains. However, existing provenance-based detection methods primarily rely on single-dimensional feature analysis, which fails to capture the dynamic and multi-dimensional patterns of modern APT attacks, resulting in insufficient detection performance. To overcome this limitation, we introduce STGAN, a model that integrates spatial-temporal graphs into host provenance graph modeling. STGAN applies temporal and spatial encoding to dynamic provenance graphs to extract temporal, spatial, and semantic features, constructing a comprehensive feature representation. This representation is further fused and enhanced using a multi-head self-attention mechanism, followed by anomaly detection. Through extensive evaluations on three widely-used provenance graph datasets, we demonstrate that our approach consistently outperforms current state-of-the-art techniques in terms of detection performance. Additionally, we contribute to the research community by releasing our datasets and code, facilitating further exploration and validation.

How can we efficiently identify the densest subgraph over relational graphs? Existing dense subgraph discovery (DSD) approaches assume that a relational graph $H$ is already derived from a heterogeneous data source and they focus on efficient discovery of the densest subgraph on the materialized $H$. Unfortunately, materializing relational graphs can be resource-intensive, which thus limits the practical usefulness of existing algorithms over large datasets. To mitigate this, we propose a novel Summary-bAsed deNsest Subgraph discovery (SANS) system. Our unique summary-based peeling algorithm forms the core of SANS. Following the peeling paradigm, it utilizes summaries of each node's neighborhood to efficiently estimate peeling coefficients and subgraph densities at each peeling iteration and thus avoids materializing the relational graph completely. Through extensive experiments, we demonstrate the efficacy and efficiency of SANS, reaching orders of magnitude speedups compared to the conventional baselines with materialization, while consistently achieving at least 95% accuracy compared to peeling algorithms based on materialization.