WWW 2025 Papers

The rapid development of large language models (LLMs), like ChatGPT, has resulted in the widespread presence of LLM-generated content on social media platforms, raising concerns about misinformation, data biases, and privacy violations, which can undermine trust in online discourse. While detecting LLM-generated content is crucial for mitigating these risks, current methods often focus on binary classification, failing to address the complexities of real-world scenarios like human-AI collaboration. To move beyond binary classification and address these challenges, we propose a new paradigm for detecting LLM-generated content. This approach introduces two novel tasks: LLM Role Recognition (LLM-RR), a multi-class classification task that identifies specific roles of LLM in content generation, and LLM Influence Measurement (LLM-IM), a regression task that quantifies the extent of LLM involvement in content creation. To support these tasks, we propose LLMDetect, a benchmark designed to evaluate detectors' performance on these new tasks. LLMDetect includes the Hybrid News Detection Corpus (HNDC) for training detectors, as well as DetectEval, a comprehensive evaluation suite that considers five distinct cross-context variations and multi-intensity variations within the same LLM role. This allows for a thorough assessment of detectors' generalization and robustness across diverse contexts. Our empirical validation of 10 baseline detection methods demonstrates that fine-tuned Pre-trained Language Model (PLM)-based models consistently outperform others on both tasks, while advanced LLMs face challenges in accurately detecting their own generated content. Our experimental results and analysis offer insights for developing more effective detection models for LLM-generated content. This research enhances the understanding of LLM-generated content and establishes a foundation for more nuanced detection methodologies.

In e-commerce platforms, profit-aware recommender systems aim to improve the platform's profits while maintaining high overall accuracy by recommending items with high profits as top-ranked items. We explore two issues faced by existing model-based profit-aware approaches (i.e., MBAs) when training recommendation models for profit enhancement. First, current MBAs tend to inaccurately infer the item ranking by the profit-based weighting scheme; the ranking of observed (i.e., purchased) items by a user is inferred without considering the user preference for each item, while all unobserved items are assumed to have an equally low ranking. Second, current MBAs train the model without employing the item ranking as ground truth; during training, the model is optimized for the preference score for each item independently rather than being directly optimized for the overall ranking of items. To tackle these issues, we propose a novel MBA that involves three key steps: (S1) defining the Current Preference incorporated with Profit (i.e., CPP) for items; (S2) classifying items through CPP; and (S3) training the model by list-wise learning-to-rank (LTR) based on CPP. Extensive experimental results using real-world platform datasets demonstrate that our approach improves accuracy by approximately 4% and profits by about 24% compared to the best-competing method.

Extreme Multi-label Classification (XMC) involves predicting a subset of relevant labels from an extremely large label space, given an input query and labels with textual features. Models developed for this problem have conventionally made use of dual encoder (DE) to embed the queries and label texts and one-vs-all (OvA) classifiers to rerank the shortlisted labels by the DE. While such methods have shown empirical success, a major drawback is their computational cost, often requiring upto 16 GPUs to train on the largest public dataset. Such a high cost is a consequence of calculating the loss over the entire label space. While shortlisting strategies have been proposed for classifiers, we aim to study such methods for the DE framework. In this work, we develop UniDEC, a loss-independent, end-to-end trainable framework which trains the DE and classifier together in a unified manner with a multi-class loss, while reducing the computational cost by 4-16x. This is done via the proposed pick-some-label (PSL) reduction, which aims to compute the loss on only a subset of positive and negative labels. These labels are carefully chosen in-batch so as to maximise their supervisory signals. Not only does the proposed framework achieve state-of-the-art results on datasets with labels in the order of millions, it is also computationally and resource efficient in achieving this performance on a single GPU. Code is provided with the submission and will be open-sourced upon acceptance.

Matrix factorization is a widely used approach for top-N recommendations and collaborative filtering. When it is implemented on implicit feedback data (such as clicks), a common heuristic is to upweight the observed interactions. This strategy has been shown to improve the performance of certain algorithms. In this paper, we conduct a systematic study of various weighting schemes and matrix factorization algorithms. Somewhat surprisingly, we find that the best performing methods, as measured by the standard (unweighted) ranking accuracy on publicly available datasets, are trained using unweighted data. This observation challenges the conventional wisdom in the literature. Nevertheless, we identify cases where weighting can be beneficial, particularly for models with lower capacity and certain regularization schemes. We also derive efficient algorithms for minimizing a number of weighted objectives which were previously unexplored due to the lack of efficient optimization techniques. Our work provides a comprehensive analysis of the interplay between weighting, regularization, and model capacity in matrix factorization for recommender systems.

Recently, research on Text-Attributed Graphs (TAGs) has gained significant attention due to the prevalence of free-text node features in real-world applications and the advancements in Large Language Models (LLMs) that bolster TAG methodologies. However, current TAG approaches face two primary challenges: (i) Heavy reliance on label information and (ii) Limited cross-domain zero/few-shot transferability. These issues constrain the scaling of both data and model size, owing to high labor costs and scaling laws, complicating the development of graph foundation models with strong transferability. In this work, we propose the GraphCLIP framework to address these challenges by learning graph foundation models with strong cross-domain zero/few-shot transferability through a self-supervised contrastive graph-summary pretraining method. Specifically, we generate and curate large-scale graph-summary pair data with the assistance of LLMs, and introduce a novel graph-summary pretraining method, combined with invariant learning, to enhance graph foundation models with strong cross-domain zero-shot transferability. For few-shot learning, we propose a novel graph prompt tuning technique aligned with our pretraining objective to mitigate catastrophic forgetting and minimize learning costs. Extensive experiments show the superiority of GraphCLIP in both zero-shot and few-shot settings, while evaluations across various downstream tasks confirm the versatility of GraphCLIP. Our code is available at: https://anonymous.4open.science/r/GraphCLIP

Link prediction in multi-layer networks is a longstanding issue that predicts missing links based on the observed structures across all layers. Existing link prediction methods in multi-layer network typically merge the multi-layer network into a single-layer network and/or perform explicit calculations using intra-layer and inter-layer similarity metrics. However, these approaches often overlook the role of coupling in multi-layer networks, specifically the shared information and latent relationships between layers, which in turn limits prediction performance. This calls the need for methods that can extract representations in a shared-latent space to enhance inter-layer information sharing and prediction performance. In this paper, we propose a novel end-to-end framework namely: Link prediction Utilizing Shared-laTent spacE Representation (LUSTER) in multi-layer networks. LUSTER consists of four key modules: the representation extractor, the latent space learner, the complementary enhancer, and the link predictor. The representation extractor focuses on learning the intra-layer representations of each layer, capturing the data characteristics within the layer. The latent space learner {extracts representations from the shared-latent space across different network layers} through adversarial training. The complementary enhancer combines the intra-layer representations and the shared-latent space representations through orthogonal fusion, providing comprehensive information. Finally, the link predictor uses the enhanced representations to predict missing links. Extensive experimental analyses demonstrate that LUSTER outperforms state-of-the-art methods for link prediction in multi-layer networks, improving the AUC metric by up to 15.87%.

The increasing prevalence of large-scale graphs poses a significant challenge for graph neural network training, attributed to their substantial computational requirements. In response, graph condensation (GC) emerges as a promising data-centric solution aiming to substitute the large graph with a small yet informative condensed graph to facilitate data-efficient GNN training. However, existing GC methods suffer from intricate optimization processes, necessitating excessive computing resources and training time. In this paper, we revisit existing GC optimization strategies and identify two pervasive issues therein: (1) various GC optimization strategies converge to coarse-grained class-level node feature matching between the original and condensed graphs; (2) existing GC methods rely on a Siamese graph network architecture that requires time-consuming bi-level optimization with iterative gradient computations. To overcome these issues, we propose a training-free GC framework termed Class-partitioned Graph Condensation (CGC), which refines the node distribution matching from the class-to-class paradigm into a novel class-to-node paradigm, transforming the GC optimization into a class partition problem which can be efficiently solved by any clustering methods. Moreover, CGC incorporates a pre-defined graph structure to enable a closed-form solution for condensed node features, eliminating the need for back-and-forth gradient descent in existing GC approaches. Extensive experiments demonstrate that CGC achieves an exceedingly efficient condensation process with advanced accuracy. Compared with the state-of-the-art GC methods, CGC condenses the Ogbn-products graph within 30 seconds, achieving a speedup ranging from $10^2 \times$ to $10^4 \times$ and increasing accuracy by up to 4.2\%.

Cross-Domain Recommendation (CDR) has been widely investi- gated for solving long-standing data sparsity problem via knowl- edge sharing across domains. In this paper, we focus on the Multi- Modal Cross-Domain Recommendation (MMCDR) problem where different items have multi-modal information while few users are overlapped across domains. MMCDR is particularly challenging in two aspects: fully exploiting diverse multi-modal information within each domain and leveraging useful knowledge transfer across domains. However, previous methods fail to cluster items with similar characteristics while filtering out inherit noises within different modalities, hurdling the model performance. What is worse, conventional CDR models primarily rely on overlapped users for domain adaptation, making them ill-equipped to handle scenarios where the majority of users are non-overlapped. To fill this gap, we propose Joint Similarity Item Exploration and Over- lapped User Guidance (SIEOUG) for solving the MMCDR problem. SIEOUG first proposes similarity item exploration module, which not only obtains pair-wise and group-wise item-item graph knowl- edge, but also reduces irrelevant noise for multi-modal modeling. Then SIEOUG proposes user-item collaborative filtering module to aggregate user/item embeddings with the attention mechanism for collaborative filtering. Finally SIEOUG proposes overlapped user guidance module with optimal user matching for knowledge sharing across domains. Our empirical study on Amazon dataset with several different tasks demonstrates that SIEOUG significantly outperforms the state-of-the-art models under the MMCDR setting

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

The rapid spread of rumor content on online micro-video platforms poses significant threats to public health and safety. However, existing Micro-Video Rumor Detection (MVRD) methods are generally black-box, which lacks transparency and makes it difficult to understand the reasoning behind classification decisions. In this work, we introduce ExMRD, a novel Explainable Micro-video Rumor Detection framework designed to generate detailed and coherent explanations for enhancing MVRD. Inspired by the powerful reasoning capacity of Chain-of-Thought (CoT), we introduce a novel inference mechanism called R^3CoT-- consisting of Refining, Retrieving, and Reasoning on MVRD. This mechanism enables Multimodal Large Language Models (MLLMs) to reorganize the original video content, retrieve domain knowledge related to rumors, and generate explainable conclusions regarding whether the micro-video contains rumor information. Instead of directly fine-tuning MLLMs for MVRD, which is computationally expensive, we propose a Small Language Reviewer (SLReviewer), which distills the outputs of R^3CoT guided MLLMs to ensure efficient and reliable predictions. Extensive experiments on three real-world benchmarks demonstrate that ExMRD significantly outperforms competitive baselines while providing high-quality rationales.

Despite the widespread adoption of large language models (LLMs) for recommendation, we demonstrate that LLMs often exhibit uncertainty in their recommendations. To ensure the trustworthy use of LLMs in generating recommendations, we emphasize the importance of assessing the reliability of recommendations generated by LLMs. We start by introducing a novel framework for estimating the predictive uncertainty to quantitatively measure the reliability of LLM-based recommendations. We further propose to decompose the predictive uncertainty into recommendation uncertainty and prompt uncertainty, enabling in-depth analyses of the primary source of uncertainty. Through extensive experiments, we (1) demonstrate predictive uncertainty effectively indicates the reliability of LLM-based recommendations, (2) investigate the origins of uncertainty with decomposed uncertainty measures, and (3) propose uncertainty-aware prompting for a lower predictive uncertainty and enhanced recommendation. Our source code and model weights are available at https://anonymous.4open.science/r/UNC_LLM_REC

Graph Neural Networks (GNNs) have expressed remarkable capability in processing graph-structured data. Recent studies have found that most GNNs rely on the homophily assumption of graphs, leading to unsatisfactory performance on heterophilous graphs. While certain methods have been developed to address heterophilous links,they lack more precise estimation of high-order relationships between nodes. This could result in the aggregation of excessive interference information during message propagation, thus degrading the representation ability of learned features. In this work, we propose a {D}isparity-induced {S}tructural {R}efinement (DSR) method that enables adaptive and selective message propagation in GNN, to enhance representation learning in heterophilous graphs. We theoretically analyze the necessity of structural refinement during message passing grounded in the derivation of error bound for node classification. To this end, we design a disparity score that combines both features and structural information at the node level, reflecting the connectivity degree of hopping neighbor nodes. Based on the disparity score, we can adjust the aggregation of neighbor nodes, thereby mitigating the impact of irrelevant information during message passing. Experimental results demonstrate that our method achieves competitive performance, mostly outperforming advanced methods on both homophilous and heterophilous datasets.

Graph Neural Networks (GNNs) have become indispensable tools in many domains, such as social network analysis, financial fraud detection, and drug discovery. Prior research primarily concentrated on improving prediction accuracy while overlooking how reliable the model predictions are. Conformal prediction on graphs emerges as a promising solution, offering statistically sound uncertainty estimates with a pre-defined coverage level. Despite the promising progress, existing works only focus on achieving model coverage guarantees without considering fairness in the coverage within different demographic groups. To bridge the gap between conformal prediction and fair coverage across different groups, we pose the fundamental question: Can fair GNNs enable the uncertainty estimates to be fairly applied across demographic groups? To answer this question, we provide a comprehensive analysis of the uncertainty estimation in fair GNNs employing various strategies. We prove theoretically that fair GNNs can enforce consistent uncertainty bounds across different demographic groups, thereby minimizing bias in uncertainty estimates. Furthermore, we conduct extensive experiments on five commonly used datasets across seven state-of-the-art fair GNN models to validate our theoretical findings. Additionally, based on the theoretical and empirical insights, we identify and analyze the key strategies from various fair GNN models that contribute to ensuring equalized uncertainty estimates. Our work estimates a solid foundation for future exploration of the practical implications and potential adjustments needed to enhance fairness in GNN applications across various domains. For reproducibility, we publish our data and code at https://anonymous.4open.science/r/EqualizedCoverage_CP-9CF8.

Feature engineering stands as a pivotal step in enhancing the performance of machine learning models, particularly with tabular data. However, traditional feature engineering methods are often time-consuming and require case-by-case domain knowledge. In addition, as machine learning systems become more common, interpretability becomes increasingly important, especially among domain experts. To this end, we propose ReaGen, an automated feature engineering (AutoFE) approach that combines the use of knowledge graphs (KGs) with large language models (LLMs) to generate interpretable features. ReaGen begins by symbolic reasoning over a knowledge graph to extract relevant information based on datasets description. Then, it uses several LLMs to iteratively generate meaningful features based on the retrieved information and the datasets description. Finally, to overcome challenges such as hallucinations and handling long contexts typical in LLMs, our model performs logical reasoning on the knowledge graph to ensure that the generated features maintain interpretability. ReaGen provides Python code for automatic feature generation and detailed explanations of feature utility. It leverages both LLM's internal knowledge and retrieved information from knowledge graphs. Extensive experiments on public datasets demonstrate that ReaGen significantly improves prediction accuracy while ensuring high interpretability through human-like explanations for each feature. This work highlights the potential of integrating large language models and knowledge graphs in feature engineering, paving the way for interpretable machine learning models.

JavaScript plays a crucial role on web. However, the inclusion of unknown, vulnerable, or malicious scripts on websites and in browser extensions and the use of browsers' developer tools often leads to undesired web content manipulations and data acquisitions. To restrict JavaScript operations on web content and data, we introduce a fine-grained, mandatory access control-based, and object-oriented permission system for browsers. With our system, web developers can define policies for sensitive web elements on their web pages to allow or deny scripts' operations on web content and data within browsers. The system substantially thwarts many web threats and attacks, and offers benefits to personal data governance. We developed a tool for automatic policy generation and demonstrated the usability and compatibility of the system in a three-month study. Our system is a reasonable and practical solution, bolstering the security and trustworthiness on the internet.

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.

The substantial data and resource consumption of training deep neural networks has rendered the large-scale training accessible only to organizations with necessary infrastructure and massive datasets. Once these models are developed, they are typically adapted to meet the diverse requirements of model owners and users through techniques like early exiting and fine-tuning. However, maintaining multiple specialized versions of the established model is inefficient and unsustainable in the long run. In response to this challenge, we propose AIM, a novel model modulation paradigm that enables a single model to exhibit diverse behaviors meeting the specific needs of stakeholders. AIM enables two key modulation modes: utility and focus modulation. The former provides model owners with dynamic control over output quality to deliver varying utility levels from the same model, the latter offers users precise control to shift model's focused features of inputs. AIM introduces a logits redistribution strategy for modulating model behaviors. It operates in a training data-agnostic and retraining-free manner by directly manipulating off-the-shelf pre-trained networks, facilitating AIM's seamless integration across diverse neural network architectures. To mathematically guarantee that our modulation achieves a precise regulation of model behavior, we establish a formal foundation grounded in the statistical properties of logits ordering via joint probability distributions. Our evaluation spans across diverse applications, including image classification, semantic segmentation, and text generation, utilizing prevalent architectures such as ResNet, SegFormer, and Llama. Experimental results confirm the efficacy of our approach, demonstrating the practicality and versatility of AIM in realizing AI model modulation. AIM provides both theoretical and system-level tools to empower a single model to meet diverse needs of both model owners and users, paving the way for scalable, accessible, and efficient AI deployment.

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.

Kubernetes is widely adopted by enterprises to enhance service availability for applications such as web services and large-scale model training, due to its advantages in managing containerized applications. As service demands increase, a single Kubernetes cluster often becomes insufficient, leading to the trend of using multiple clusters to improve service scalability. However, achieving efficient cross-cluster communication poses significant challenges due to the need for low latency, high throughput, and strong robustness. Existing methods for cross-cluster communication either employ a centralized control plane, which becomes a communication bottleneck, or use numerous service-bound proxies, leading to increased management complexity and possibly compromised robustness in cross-cluster communication. To address the above challenges, we introduce X-ClusterLink, a framework designed for efficient cross-cluster communication in multi-Kubernetes clusters. X-ClusterLink first employs broker clusters to ensure low-latency cross-cluster synchronization. Then, it aggregates multiple containerized gateways to enhance throughput and leverages eXpress Data Path (XDP) for advanced packet processing, thereby accelerating traffic forwarding. Finally, it incorporates Bucket-Based Consistent ECMP to facilitate seamless failover and enhance robustness. Experimental results demonstrate that X-ClusterLink significantly improves cross-cluster communication efficiency, increasing cross-cluster forwarding bandwidth by 3.1 $\times$ compared to existing solutions.