WWW 2025 论文列表

With the escalating privacy concerns in recommender systems, attribute unlearning has drawn widespread attention as an effective approach against attribute inference attacks. This approach focuses on unlearning users' privacy attributes to reduce the performance of attackers while preserving the overall effectiveness of recommendation. Current research attempts to achieve attribute unlearning through adversarial training and distribution alignment in the statistic setting. However, these methods often struggle in dynamic real-world environments, particularly when considering scenarios where unlearning requests are frequently updated. In this paper, we first identify three main challenges of current methods in dynamic environments, i.e., irreversible operation, low efficiency, and unsatisfied recommendation preservation. To overcome these challenges, we propose a Pluggable Attribute Unlearning framework, PAU. Upon receiving an unlearning request, PAU plugs an additional erasure module into the original model to achieve unlearning. This module can perform a reverse operation if the request is later withdrawn. To enhance the efficiency of unlearning, we introduce rate distortion theory and reduce the attack performance by maximizing the encoded bits required for users' embedding within the same class of the unlearned attribute and minimizing those for different classes. We further preserve recommendation performance by constraining the compactness of the user embedding space using an adjustable flooding parameter/around a reasonable flooding level. Extensive experiments conducted on four real-world datasets and three mainstream recommendation models demonstrate the effectiveness of our proposed framework.

Processing long contexts presents a significant challenge for large language models (LLMs). While recent advancements allow LLMs to handle much longer contexts than before (e.g., 32K or 128K tokens), it is computationally expensive and can still be insufficient for many applications. Retrieval-Augmented Generation (RAG) is considered as a promising strategy to address this problem. However, conventional RAG methods face inherent limitations because of two underlying requirements: 1) explicitly-stated queries, and 2) well-structured knowledge. These conditions, however, do not hold in general long-context processing tasks. In this work, we propose HawkRAG, a novel RAG framework empowered by global memory-augmented retrieval. HawkRAG features a dual-system architecture. The name HawkRAG is inspired by the way a hawk glides high in the sky to observe the land, allowing it to spot and target prey with precision from a broad vantage point. First, it employs a light but long-range system to create a global memory of the long context. Once a task is presented, it generates draft answers, providing useful clues for the retrieval tools to locate relevant information within the long context. Second, it leverages an expensive but expressive system, which generates the final answer based on the retrieved information. Building upon this fundamental framework, we realize the memory module in the form of KV compression, and reinforce its memorization and cluing capacity from the Generation quality's Feedback (a.k.a. RLGF). In our experiments, HawkRAG achieves superior performances across a variety of long-context evaluation tasks, not only complex scenarios where traditional RAG methods struggle, but also simpler ones where RAG is typically applied.

Graphs are widely used to model complex systems across various domains, including social networks and biological systems. A key task in graph analysis is identifying recurring structural patterns, known as graphlets, which capture connectivity among a fixed-size subset of nodes. While graphlets have been extensively applied in tasks such as measuring graph similarity and identifying communities, conventional graphlets focus only on direct connections between nodes. This limitation overlooks potential insights from more distant relationships within the graph structure. In this paper, we introduce (𝑑, 𝑠)-graphlets, a generalization of size-𝑠 graphlets that incorporates indirect connections between nodes up to distance 𝑑. This new formulation provides a more fine-grained and comprehensive understanding of local graph structures. To efficiently count (𝑑, 𝑠)-graphlets in a graph, we present EDGE, an exact counting algorithm that employs optimized combinatorial techniques to significantly reduce computational complexity compared to naive enumeration. Our empirical analysis across diverse real-world datasets demonstrates that (𝑑, 𝑠)-graphlets provide superior graph characterization, outperforming conventional graphlets in the graph clustering task. Moreover, our case studies show that (𝑑, 𝑠)-graphlets uncover non-trivial insights that would remain undiscovered when using conventional graphlets.

With the explosive growth of online service platforms, an increasing number of people and enterprises are undertaking personal and professional tasks online. In real applications such as trip planning and online marketing, planning sessions for a sequence of activities or services will enable social users to receive the optimal services, improving their experience and reducing the cost of their activities. These online platforms are heterogeneous, including different types of services with different attributes. However, the problem of session planning over heterogeneous platforms has not been studied so far. In this paper, we propose a Motivation-Aware Session Planning (MASP) framework for session planning over heterogeneous social platforms. Specifically, we first propose a novel HeterBERT model to handle the heterogeneity of items at both type and attribute levels. Then, we propose to predict user preference using the motivations behind user activities. Finally, we propose an algorithm together with its optimisations for efficient session generation. The extensive tests prove the high effectiveness and efficiency of MASP.

Traffic prediction plays a pivotal role in intelligent transportation systems. Most existing studies only predict traffic flow for a specific time period based on traffic data from a short period, such as an hour, overlooking the influence of periodicity present in traffic data. Moreover, most of the existing advanced methods rely on manually constructed spatio-temporal graphs for joint modeling, or use pure spatial and pure temporal modules to separately model spatial and temporal features, which limits the learning of complex spatio-temporal patterns in traffic data due to structural inadequacies in the model. To address these issues, we propose a novel approach by constructing a learnable long-range spatio-temporal graph, which can better capture complex patterns in traffic data. We introduce a new model, LLGformer, which improves upon traditional Transformer-style models, facilitating more efficient learning of traffic flow data by integrating long-range historical information. Leveraging attention mechanisms on a spatiotemporal graph enables direct interaction of information across different time slices and locations. Additionally, we propose two optimization strategies to further boost the speed of training and inference. Extensive experiments on four real-world datasets show that the new model significantly outperforms state-of-the-art methods.

Machine unlearning is an emerging field that selectively removes specific data samples from a trained model. This capability is crucial for addressing privacy concerns, complying with data protection regulations, and correcting errors or biases introduced by certain data. Unlike traditional machine learning, where models are typically static once trained, machine unlearning facilitates dynamic updates that enable the model to "forget" information without requiring complete retraining from scratch. There are various machine unlearning methods, some of which are more time-efficient when data removal requests are fewer. To decrease the execution time of such machine unlearning methods, we aim to reduce the size of data removal requests based on the fundamental assumption that the removal of certain data would not result in a distinguishable retrained model. We first propose the concept of unnecessary unlearning, which indicates that the model would not alter noticeably after removing some data points. Subsequently, we review existing solutions that can be used to solve our problem. We highlight their limitations in adaptability to different unlearning scenarios and their reliance on manually selected parameters. We consequently put forward FUNU, a method to identify data points that lead to unnecessary unlearning. FUNU circumvents the limitations of existing solutions. The idea is to discover data points within the removal requests that have similar neighbors in the remaining dataset. We utilize a reference model to set parameters for finding neighbors, inspired from the area of model memorization. We provide a theoretical analysis of the privacy guarantee offered by FUNU and conduct extensive experiments to validate its efficacy.

Advanced Persistent Threats (APTs) are continuously evolving, leveraging their stealthiness and persistence to put increasing pressure on current provenance-based Intrusion Detection Systems (IDS). This evolution exposes several critical issues: (1) The dense interaction between malicious and benign nodes within provenance graphs introduces neighbor noise, hindering effective detection; (2) The complex prediction mechanisms of existing APTs detection models lead to the insufficient utilization of prior knowledge embedded in the data; (3) The high computational cost makes detection impractical. To address these challenges, we propose Vodka, a lightweight threat detection system built on a knowledge distillation framework, capable of node-level detection within audit log provenance graphs. Specifically, Vodka applies graph Laplacian regularization to reduce neighbor noise, obtaining smoothed and denoised graph signals. Subsequently, Vodka employs a teacher model based on GNNs to extract knowledge, which is then distilled into a lightweight student model. The student model is designed as a trainable combination of a feature transformation module and a personalized PageRank random walk label propagation module, with the former capturing feature knowledge and the latter learning label and structural knowledge. After distillation, the student model benefits from the knowledge of the teacher model to perform precise threat detection. Finally, Vodka reconstructs attack paths from anomalous nodes, providing insight into the attackers' strategies. We evaluate Vodka through extensive experiments on three public datasets and compare its performance against several state-of-the-art IDS solutions. The results demonstrate that Vodka achieves outstanding detection accuracy across all scenarios and the detection time is 1.4 to 5.2 times faster than the current state-of-the-art methods.

Retrieval-augmented generation (RAG) is a well-suited technique for retrieving privacy-sensitive Electronic Health Records (EHR). It can serve as a key module of the healthcare copilot, helping reduce misdiagnosis for healthcare practitioners and patients. However, the diagnostic accuracy and specificity of existing heuristic-based RAG models used in the medical domain are inadequate, particularly for diseases with similar manifestations. This paper proposes MedRAG, a RAG model enhanced by knowledge graph (KG)-elicited reasoning for the medical domain that retrieves diagnosis and treatment recommendations based on manifestations. MedRAG systematically constructs a comprehensive four-tier hierarchical diagnostic KG encompassing critical diagnostic differences of various diseases. These differences are dynamically integrated with similar EHRs retrieved from an EHR database, and reasoned within a large language model. This process enables more accurate and specific decision support, while also proactively providing follow-up questions to enhance personalized medical decision-making. MedRAG is evaluated on both a public dataset DDXPlus and a private chronic pain diagnostic dataset (CPDD) collected from our cooperating hospital, and its performance is compared against various existing RAG methods. Experimental results show that, leveraging the information integration and relational abilities of the KG, our MedRAG provides more specific diagnostic insights and outperforms state-of-the-art models in reducing misdiagnosis rates. Our code will be available at https://github.com/username00-c/MedRAG.git.

Link prediction methods predict missing facts in incomplete knowledge graphs, often using embeddings to enhance scalability. However, embeddings complicate explainability, which is crucial for users' understanding of inferences in many domains. Methods emerged to explain predictions by identifying supporting portions of knowledge. To evaluate explanations from a user perspective, they can be compared to those in benchmarks, though they are limited to simplistic graphs. In contrast, user studies on forward simulatability variation measure how explanations improve predictability, i.e., the user ability to predict the results of inferences, which is key to trust. However, user studies face scalability and reproducibility issues on large graphs. Recognizing these gaps, we propose LP-DIXIT to algorithmically evaluate explanations of link predictions by determining forward simulatability variation and adopting large language models to mimic users, as is done in other domains, e.g., in evaluating other approaches on language related tasks. We experimentally prove that LP-DIXIT evaluates as effective explanations those in benchmarks, and we adopt it to compare state-of-the-art explanation methods.

Recently, many efforts have been devoted to building effective sequential recommenders. Despite their effectiveness, these methods typically develop a single model to serve all users. However, our empirical studies reveal that different sequential encoders have intrinsic architectural biases and tend to focus on specific behavioral patterns, \ie particular frequency range of user behavior sequences. For example, the Self-Attention module is essentially a low-pass filter, focusing on low-frequency information while neglecting the high-frequency details. This evidently limits their ability to capture diverse user patterns, leading to suboptimal recommendations. To tackle this problem, we present FamouSRec, a Frequency-Augmented mixture-of-Heterogeneous-Experts Framework for personalized recommendations. Our approach builds an MoE-based recommender system, integrating the strengths of various experts to achieve diversified user modeling. For developing the MoE framework, as the key to our approach, we instantiate experts with various model architectures, aiming to leverage their inherent architectural biases and capture diverse behavioral patterns. For selecting appropriate experts to serve individuals, we introduce a frequency-augmented router. It first identifies frequency components in user behavior sequences that are suited for expert encoding, and then conducts customized routing based on the informativeness of these components. Building on this framework, we further propose two novel contrastive tasks to enhance expert specialization and alignment, thus further improving modeling efficacy and enabling robust recommendations. Extensive experiments on five real-world datasets demonstrate the effectiveness of our approach.

Personalized learner modeling uses learners' historical behavior data to diagnose their cognitive abilities, a process known as Cognitive Diagnosis (CD) in the literature. This is a fundamental yet crucial task in web-based learning services, such as learning resource recommendation and adaptive testing. Previously, researchers discovered that models improperly correlate learners' abilities with their sensitive attributes, resulting in unfair diagnoses for learners from different sensitive groups (e.g., gender, region). Given the input of sensitive attributes, researchers proposed decorrelating these attributes from the modeling process, demonstrating improved fairness results. However, privacy concerns make collecting sensitive attributes impractical. This challenge is compounded by the presence of multiple sensitive attributes, making fairness improvement under any of them difficult. In this paper, we explore how to achieve fair personalized learner modeling without relying on any sensitive attribute input. Specifically, we first introduce a novel fairness objective tailored for personalized learner modeling. We then propose a max-min strategy that facilitates both potential sensitive information inference and fair CD modeling. In the max step, we propose a pseudo-label inference method based on maximizing the designed fairness objective. Given these pseudo-labels, the min step involves retraining a fair CD model by minimizing the designed objective. Additionally, we provide a theoretical guarantee that implementing our proposed framework reduces the upper bound of fairness generalization error. Extensive experiments demonstrate that the proposed framework significantly outperforms existing methods in terms of fairness and accuracy. Our code is available at https://anonymous.4open.science/r/FairWISA-40C6/.

Graph self-supervised learning has gained significant attention recently. However, many existing approaches heavily depend on perturbations, and inappropriate perturbations may corrupt the graph’s inherent information. The Vector Quantized Variational Autoencoder (VQ-VAE) is a powerful autoencoder extensively used in fields such as computer vision; however, its application to graph data remains underexplored. In this paper, we provide an empirical analysis of vector quantization in the context of graph autoencoders, demonstrating its significant enhancement of the model's capacity to capture graph topology. Furthermore, we identify two key challenges associated with vector quantization when applying in graph data: codebook underutilization and codebook space sparsity. For the first challenge, we propose an annealing-based encoding strategy that promotes broad code utilization in the early stages of training, gradually shifting focus toward the most effective codes as training progresses. For the second challenge, we introduce a hierarchical two-layer codebook that captures relationships between embeddings through clustering. The second layer codebook links similar codes, encouraging the model to learn closer embeddings for nodes with similar features and structural topology in the graph. Our proposed model outperforms 16 representative baseline methods in self-supervised link prediction and node classification tasks across multiple datasets. Our implementation is available at https://anonymous.4open.science/r/hqa-gae-D2F4.

With the development of multi-modality data modeling techniques, recent recommender systems use not only textual data and user-item interactions but also multi-modality data such as images to improve their performances. Existing methods typically adopt cross-modal pairwise alignment strategies to alleviate the gap between modalities. Nevertheless, this alignment paradigm has limitations on explainability, consistency, and expansibility, which may only achieve suboptimal performances. In this paper, we propose a novel Explainable generative multi-modality Alignment method for transferable Recommender systems, i.e., EARec. Specifically, we design a two-stage pipeline to achieve unified multi-modality alignment of items and the sequential recommendation task, respectively. In the first phase, we present a generation task that parallel aligns each modality from multiple source domains to an anchor with explainable meaning. Three modality features share the same anchor to achieve a consistent alignment direction. Additionally, we incorporate behavior-related information as an independent modality into the alignment framework, establishing a bridge that promotes the alignment between multi-modalities and behavior. In the second stage, we composite the aligned modality encoders into a unified one and then transfer it to the target domain to enhance sequential recommendation. The pipeline that adopts parallel multi-modal alignment and composition shows flexibility and scalability for incorporating new modalities. Experimental results on multiple public datasets demonstrate the superiority of EARec over multi-modality recommendation baselines and further analysis indicates the explainability of generative alignment.

Platforms design the forms of presentation by which sellers are shown to the buyers. This design not only shapes the buyers' experience but also leads to different market equilibria or dynamics. One component in this design is through the platform's mediation of the search frictions experienced by the buyers for different sellers. We take a model of monopolistic competition and show that, on one hand, when all sellers have the same inspection costs, the market sees no stable price as the sellers always have incentive to undercut each other, and, on the other hand, the platform may stabilize the price by giving prominence to one seller chosen by a carefully designed mechanism. This calls to mind Amazon's Buy Box design. We study natural mechanisms for choosing the prominent seller, characterize the range of equilibrium prices implementable by them, and find, somewhat counterintuitively, that in certain scenarios the buyers' surplus improves as the search friction increases.

Graph prompt tuning has achieved significant success for its ability to effectively adapt pre-trained graph neural networks to various downstream tasks. However, the pre-trained models may learn discriminatory representation due to the inherent prejudice in graph-structured data. Existing graph prompt tuning overlooks such unfairness, leading to biased outputs towards certain demographic groups determined by sensitive attributes such as gender, age, and political ideology. To overcome this limitation, we propose a fairness-aware graph prompt tuning method to promote fairness while enhancing the generality of any pre-trained GNNs (named FPrompt). FPrompt introduces hybrid graph prompts to augment counterfactual data while aligning the pre-training and downstream tasks. It also applies edge modification to increase sensitivity heterophily. We provide a two-fold theoretical analysis: first, we demonstrate that FPrompt possesses universal capabilities in handling pre-trained GNN models across various pre-training strategies, ensuring its adaptability in different scenarios. Second, we show that FPrompt effectively reduces the upper bound of generalized statistical parity, thereby mitigating the bias of pre-trained models. Extensive experiments demonstrate that FPrompt outperforms baseline models in both accuracy and fairness (~$33$\%) on benchmark datasets. Additionally, we introduce a new benchmark for transferable evaluation, showing that FPrompt achieves state-of-the-art generalization performance.

Cross-Domain Sequential Recommendation (CDSR) has recently gained attention for countering data sparsity by transferring knowledge across domains. A common approach merges domain-specific sequences into cross-domain sequences, serving as bridges that enable mutual enhancement between domains. One key challenge is to correctly extract the effective shared knowledge among these sequences and appropriately transfer it. Most existing works directly transfer unfiltered cross-domain knowledge rather than extracting domain-invariant components and adaptively integrating them into domain-specific modelings. Another challenge lies in aligning the domain-specific and cross-domain sequences. Existing methods align these sequences based on timestamps, but this approach can cause prediction mismatches when the current tokens and their targets belong to different domains. In such cases, the domain-specific knowledge carried by the current tokens may degrade performance. To address these challenges, we propose the A-B-Cross-to-Invariant Learning Recommender (\textbf{ABXI}). Specifically, leveraging LoRA's effectiveness for efficient adaptation as supported by numerous studies, our model incorporates two types of LoRAs to facilitate the adaptation process. First, all sequences are processed through a shared encoder that employs a domain LoRA for each sequence, thereby preserving unique domain characteristics. Next, we introduce an invariant projector that extracts domain-invariant interests from cross-domain representations, utilizing an invariant LoRA as well to adapt these interests into recommendations in each specific domain. Besides, to avoid prediction mismatches, all domain-specific sequences are re-aligned to match the domains of the cross-domain ground truths. Experimental results on three datasets demonstrate that our approach achieves better results than other CDSR counterparts, with an average improvement of 17.30\% in HR@10 and 18.65\% in NDCG@10.

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.