WWW 2025 Papers

Many ethical issues in machine learning are connected to the training data. Online data markets are an important source of training data, facilitating both production and distribution. Recently, a trend has emerged of for-profit “ethical” participants in online data markets. This trend raises a fascinating question: Can online data markets sustainably and efficiently address ethical issues in the broader machine-learning economy? In this work, we study this question in a stylized model of an online data market. We investigate the effects of intervening in the data market to achieve balanced training-data production. The model reveals the crucial role of market conditions. Under some conditions, an intervention can drive the data producers out of the market, so that the cost of fairness is maximal. Yet, under other conditions, the cost of fairness can vanish (as a fraction of overall welfare) as the market grows. Our results suggest that “ethical” online data markets can be economically feasible under favorable market conditions, and motivate more work to consider the role of data production and distribution in mediating the impacts of ethical interventions.

Dynamic link prediction aims to predict the future links on dynamic graphs, which can be applied to wide scenarios such as recommender systems and social networks on the World Wide Web. Existing methods mainly (1) focus on the in-graph learning, which cannot generalize to graphs unobserved during training; or (2) achieve the cross-graph predictions in a many-many mechanism by training on multiple graphs across various domains, which results in a large computational cost. In this paper, we propose a cross-graph dynamic link predictor named CrossDyG, which achieves the cross-graph transferability in a one-many mechanism which trains on one single source graph and test on different target graphs. Specifically, we provide causal and empirical analysis on the structural bias caused by the graph-specific structural characteristics in cross-graph predictions. Then, we conduct deconfounded training to learn the universal network evolution pattern from one single source graph during training. Finally, we apply the causal intervention to leverage the graph-specific structural characteristics of each target graph during inference. Extensive experiments conducted on three benchmark data of dynamic graphs demonstrate that CrossDyG outperforms the state-of-the-art baselines by up to 11.01% and 17.02% in terms of AP and AUC, respectively. In addition, the improvements are especially significant when training on small source graphs. The implementation of our approach is available in https://anonymous.4open.science/r/CrossDyG-8B70.

The World Wide Web thrives on intelligent services that depend heavily on accurate time series forecasting to navigate dynamic and evolving environments. Due to the partially-observed nature of real world, exclusively focusing on the target of interest, so-called \textit{endogenous variables}, is insufficient for accurate forecasting, especially in web systems that are susceptible to external influences. Thus, utilizing \textit{exogenous variables} to harness external information, i.e., forecasting with exogenous variable (FEV), is imperative. Nevertheless, as the external environment is complex and ever-evolving, inadequately capturing external influences can even lead to learning spurious correlations and invalid prediction. Fortunately, recent studies have demonstrated that large language models (LLMs) exhibit exceptional recognition capabilities across open real-world systems, including a deep understanding of exogenous environments. However, it is difficult to directly apply LLMs for FEV due to challenges of task activation, exogenous knowledge extraction, and feature space alignment. In this work, we devise ExoLLM, an \underline{LLM}-driven method to sufficiently utilize \underline{Exo}genous variables for time series forecasting. We begin by Meta-task Instruction to activate the knowledge transfer of LLM from natural language processing to FEV. To comprehensively understand the intricate and hierarchical influences of exogenous variables, we propose Multi-grained Prompts, encompassing diverse external influences, including natural attributes, trend correlations, and period relationships between two types of variables. Additionally, a Dual TS-Text Attention is devised to bridge the feature gap between text and numeric data in LLM. Evaluation on real-world datasets demonstrates ExoLLM's superiority in exploiting exogenous information for forecasting with open-world language knowledge. Code is available at \url{https://anonymous.4open.science/r/ExoLLM}.

Being able to query online public knowledge graphs such as Wikidata or DBpedia is extremely valuable. However, these queries can be interrupted due to the fair use policies enforced by SPARQL endpoint providers, leading to incomplete results. While these policies help maintain responsiveness for public SPARQL endpoints, they compromise the completeness of query results, limiting the feasibility of various downstream tasks. Ideally, we shouldn’t have to choose between completeness and responsiveness. To address this issue, we introduce the concept of SPARQL continuation queries. When a SPARQL endpoint interrupts a query, it returns partial results along with a SPARQL continuation query to retrieve the remaining results. If the continuation query is also interrupted, the process repeats, generating further continuation queries until the complete results are obtained. In our experimention, we show that our continuation server Passage ensures completeness and responsiveness with performances in execution time similar to BlazeGraph.

Existing domain-specific Large Language Models (LLMs) are typically developed by fine-tuning general-purposed LLMs with large-scale domain-specific corpora. However, training on large-scale corpora often fails to effectively organize domain knowledge of LLMs, leading to fragmented understanding. Inspired by how humans connect concepts and organize knowledge through mind maps, we aim to emulate this approach by using ontology with hierarchical conceptual knowledge to reorganize LLM's domain knowledge. From this perspective, we propose an ontology-driven self-training framework called OntoTune, which aims to align LLMs with ontology through in-context learning, enabling the generation of responses guided by the ontology. We leverage in-context learning to identify whether the LLM has acquired the specific concept's ontology knowledge, and select the entries not yet mastered by LLM as the training set to further align the LLM with ontology. Compare to existing domain LLMs based on newly collected large-scale domain-specific corpora, our OntoTune, which relies on the existing, long-term developed ontology and LLM itself, significantly reduces data maintenance costs and offers improved generalization ability. We conduct our study in the medical domain to evaluate the effectiveness of OntoTune, utilizing a standardized medical ontology, SNOMED CT as our ontology source. Experimental results demonstrate that OntoTune achieves state-of-the-art performance in both in-ontology task hypernym discovery and out-of-ontology task medical domain QA. Moreover, compared to the latest direct ontology injection method TaxoLLaMA, our OntoTune better preserves original knowledge of LLM.

Tags play a critical role in enhancing product discoverability, optimizing search results, and enriching recommendation systems on e-commerce platforms. Despite the recent advancements in large language models (LLMs), which have shown proficiency in processing and understanding textual information, their application in tag generation remains an under-explored yet complex challenge. To this end, we introduce a novel method for automatic product tagging using LLMs to create behavior-enhanced tags (BETags). Specifically, our approach begins by generating base tags using an LLM. These base tags are then refined into BETags by incorporating user behavior data. This method aligns the tags with users' actual browsing and purchasing behavior, enhancing the accuracy and relevance of tags to user preferences. By personalizing the base tags with user behavior data, BETags are able to capture deeper behavioral insights, which is essential for understanding nuanced user interests and preferences in e-commerce environments. Moreover, since BETags are generated offline, they do not impose real-time computational overhead and can be seamlessly integrated into downstream tasks commonly associated with recommendation systems and search optimization. Our evaluation of BETag across three datasets--- Amazon (Scientific), MovieLens-1M, and FreshFood---shows that our approach significantly outperforms both human-annotated tags and other automated methods. These results highlight BETag as a scalable and efficient solution for personalized automated tagging, advancing e-commerce platforms by creating more tailored and engaging user experiences.

As large language models (LLMs) are gaining increasing popularity across a wide range of web applications, it is of great importance to optimize service-level objectives (SLOs) for LLM inference services to enhance user satisfaction and improve the competitiveness of cloud vendors. In this paper, we observe that adjusting the parameters of LLM inference engines can improve service performance, and the optimal parameter configurations of different services are different. Therefore, we propose SCOOT, an automatic performance tuning system to optimize SLOs for each LLM inference service by tuning the parameters of the inference engine. SCOOT jointly exploits single-objective and multiple-objective Bayesian optimization (BO) techniques to handle various optimization objectives via exploration and exploitation. Moreover, SCOOT prunes the search space with known constraints and adopts a random forest to learn hidden constraints during the tuning process to mitigate invalid exploration. To improve the tuning efficiency, SCOOT utilizes the parallel suggestion to accelerate the tuning process. Extensive experiments demonstrate that SCOOT considerably outperforms existing tuning techniques in SLO optimization while greatly improving the tuning efficiency. SCOOT is universally applicable to various LLM inference engines and is easily expandable to new parameters. Currently, SCOOT has already been implemented in the production environment of a leading international technology company.

Ranking is an essential and practical task on dynamic graphs, which aims to prioritize future interaction candidates for given queries. While existing solutions achieve promising ranking performance, they leverage a single listwise loss to jointly optimize candidate sets, which leads to the gradient vanishing issue; and they employ neural networks to model complex temporal structures within a shared latent space, which fails to accurately capture multi-scale temporal patterns due to the frequency aliasing issue. To address these issues, we propose BandRank, a novel and robust band-pass disentangled ranking approach for dynamic graphs in the frequency domain. Concretely, we propose a band-pass disentangled representation (BPDR) approach, which disentangles complex temporal structures into multiple frequency bands and employs non-shared frequency-enhanced multilayer perceptrons (MLPs) to model each band independently. We prove that our BPDR approach ensures effective multi-scale learning for temporal structures by demonstrating its multi-scale global convolution property. Besides, we design a robust Harmonic Ranking (HR) loss to jointly optimize candidate sets and continuously track comparisons between real and virtual candidates, where we theoretically guarantee its ability to alleviate the gradient vanishing issue. Extensive experimental results show that our BandRank achieves an average improvement of 21.31% against eight baselines while demonstrating superior robustness across different learning scenarios.

Traditional recommender systems rely on high-dimensional (latent) embeddings for modeling user-item interactions, often resulting in opaque representations that lack interpretability. Moreover, these systems offer limited control to users over their recommendations. Inspired by recent work, we introduce TExtuAl Representations for Scrutable recommendations (TEARS) to address these challenges. Instead of representing a user’s interests through latent embed- dings, TEARS encodes them in natural text, providing transparency and allowing users to edit them. To encode such preferences, we use modern LLMs to generate high-quality user summaries which we find uniquely capture user preferences. Using these summaries we take a hybrid approach where we use an optimal transport procedure to align the summaries’ representations with the repre- sentation of a standard VAE for collaborative filtering. We find this approach can surpass the performance of the three popular VAE models while providing user-controllable recommendations. We further analyze the controllability of TEARS through three simu- lated user tasks to evaluate the effectiveness of user edits on their summaries. Our code and all user-summaries can be seen in an anonymized repository.

Conversational recommender systems (CRS) aim to provide personalized recommendations via interactive dialogues with users. While large language models (LLMs) enhance CRS with their superior understanding of context-based user preferences, they typically struggle to leverage behavioral data, which has proven to be the key for classical collaborative filtering approaches. For this reason, we propose CRAG—Collaborative Retrieval Augmented Generation for LLM-based CRS. To the best of our knowledge, CRAG is the first approach that combines state-of-the-art LLMs with collaborative filtering for conversational recommendations. Our experiments on two publicly available conversational datasets in the movie domain, i.e., a refined Reddit dataset as well as the Redial dataset, demonstrate the superior item coverage and recommendation performance of CRAG, compared to several CRS baselines. Moreover, we observe that the improvements are mainly due to better recommendation accuracy on recently released movies. The code is anonymously available at: https://anonymous.4open.science/r/CRAG-8CBE.

Machine learning (ML) models are widely deployed on edge nodes, such as mobile phones and edge servers, to power a wide range of AI applications over the web. Ensuring the integrity of these edge models is paramount, as they are subject to corruption caused by software/hardware exceptions and malicious tampering, which may undermine model performance, incur economic losses, and pose health risks. Existing data integrity mechanisms designed for files stored on disks cannot properly verify the integrity of models running in GPUs or mitigate the new integrity threats against edge models. This paper proposes EdgeThemis, a novel mechanism for verifying the integrity of edge models through sentinel verification. To enable verifiability for a model $M$, EdgeThemis embeds a sentinel backdoor and a verification module into $M$. Then, a challenger can send verification requests to the edge node hosting $M$ to verify its integrity. Next, the sentinel activates the verification module to generate a unique integrity proof tied to the identity of the edge node for verification. Finally, the challenger can verify the integrity proof to detect model corruption. Theoretical analysis proves that EdgeThemis can properly mitigate potential integrity threats against edge models. Experiments demonstrate that EdgeThemis achieves a verification accuracy of 100.00\% across various models and different types of model corruption with robustness against replay attacks, theft attacks, and replacement attacks.

The spread of fake news negatively impacts individuals and is regarded as a significant social challenge that needs to be addressed. A number of algorithmic and insightful features have been identified for detecting fake news. However, with the recent LLMs and their advanced generation capabilities, many of the detectable features (e.g., style-conversion attacks) can be altered, making it more challenging to distinguish from real news. This study proposes adversarial style augmentation, AdSyle, to train a fake news detector that remains robust against various style-conversion attacks.Our model's key mechanism is the careful use of LLMs to automatically generate a diverse yet coherent range of style-conversion attack prompts. This improves the generation of prompts that are particularly difficult for the detector to handle. Experiments show that our augmentation strategy improves robustness and detection performance when tested on fake news benchmark datasets.

Large language models (LLMs) integrated with retrieval-augmented generation (RAG) systems enhance accuracy by accessing external knowledge database. However, recent studies have exposed RAG's vulnerability to poisoning attacks, where an attacker inject poisoned texts into the knowledge database, leading to attacker-desired responses. Existing defenses, primarily focused on inference-time mitigation, have proven inadequate against sophisticated attacks. In this paper, we present the first traceback system in RAG, RAGForensics, which traces poisoned texts from the knowledge database. RAGForensics narrows the space of potentially poisoned texts and accurately identifies them without requiring access to model gradients, a common challenge in RAG systems. Our empirical evaluation on multiple datasets demonstrates RAGForensics's effectiveness against state-of-the-art and adaptive poisoning attacks. This work pioneers the exploration of poisoned texts traceback in RAG systems, offering a practical and promising approach to securing them against poisoning attacks.

With the rapid increase in paper submissions to academic conferences, the need for automated and accurate paper-reviewer matching is more critical than ever. Previous efforts in this area have considered various factors to assess the relevance of a reviewer's expertise to a paper, such as the semantic similarity, shared topics, and citation connections between the paper and the reviewer's previous works. However, most of these studies focus on only one factor, resulting in an incomplete evaluation of the paper-reviewer relevance. To address this issue, we propose a unified model for paper-reviewer matching that jointly considers semantic, topic, and citation factors. To be specific, during training, we instruction-tune a contextualized language model shared across all factors to capture their commonalities and characteristics; during inference, we chain the three factors to enable step-by-step, coarse-to-fine search for qualified reviewers given a submission. Experiments on four datasets (one of which is newly contributed by us) spanning various fields such as machine learning, computer vision, information retrieval, and data mining consistently demonstrate the effectiveness of our proposed Chain-of-Factors model in comparison with state-of-the-art paper-reviewer matching methods and scientific pre-trained language models.

Federated Recommendation System (FRS) usually offers recommendation services for users while keeping their data locally to ensure privacy. Currently, most FRS literature assumes that fixed users participate in federated training with personal IoT devices (e.g., mobile phones and PC). However, users may come incrementally, and it is unfeasible to retrain the whole FRS with the new participating user due to the expensive training overheads and the negligible global knowledge gain brought by a small number of new users. To guarantee the quality service for these new users, we take a dive into the federated recommendation for cold-start users, a novel scenario where the new participating users can directly achieve a promising recommendation without overall training with all participating users by leveraging both transferred knowledge from the converged warm clients and the knowledge learned from the local data. Nevertheless, how to efficiently transfer knowledge from warm clients remains controversial. On the one hand, cold clients may introduce new sparse items, causing a distribution shift from the item embedding converged on warm clients. On the other hand, the user information from warm clients is required to match cold users for a collaborative recommendation, but directly sharing user information is a violation of privacy and unacceptable. To tackle these challenges, we propose an efficient and privacy-enhanced federated recommendation for cold-start users (FR-CSU) that each client can adaptively transfer both user and item knowledge from warm clients separately and implement recommendations with local and transferred knowledge fusion. Specifically, each cold client will train a mapping function locally to transfer the aligned item embedding. Meanwhile, warm clients will maintain a user prototype network in a FedAvg manner that provides privacy-friendly yet effective user information for cold users. Finally, a linear function system will fuse the transferred and local knowledge to improve the recommendation. Extensive experiments show that FR-CSU achieves superior performance compared to state-of-the-art methods.

Incorporating a critiquing component into recommender applications facilitates the enhancement of user perception. Typically, critique-able recommender systems adapt the model parameters and update the recommendation list in real-time through the analysis of user critiquing keyphrases in the inference phase. The current critiquing methods necessitate the designation of a dedicated recommendation model to estimate user relevance to the critiquing keyphrase during the training phase preceding the recommendations update. This paradigm restricts the applicable scenarios and reduces the potential for keyphrase exploitation. Furthermore, these approaches ignore the issue of catastrophic forgetting caused by continuous modification of model parameters in multi-step critiquing. Thus, we present a general $\textbf{R}epresentative$ ${\textbf{I}tems}$ ${\textbf{S}ampling}$ $Framework$ $for$ $\textbf{C}ritiquing$ $on$ $Knowledge$ $Graph$ ${Recommendation}$ (RISC) implemented as a plug-in, which offers a new paradigm for critiquing in mainstream recommendation scenarios. RISC leverages the knowledge graph to sample important representative items as a hinge to expand and convey information from user critiquing, indirectly estimating the relevance of the user to the critiquing keyphrase. Consequently, the necessity for specialized user-keyphrase correlation modules is eliminated with respect to a variety of knowledge graph recommendation models. Moreover, we propose a ${\textbf{W}eight}$ $\textbf{E}xperience$ $\textbf{R}eplay$ (WER) approach based on KG to mitigate catastrophic forgetting by reinforcing the user's prior preferences during the inference phase. Our extensive experimental findings on three real-world datasets and three knowledge graph recommendation methods illustrate that RISC with WER can be effectively integrated into knowledge graph recommendation models to efficiently utilize user critiquing for refining recommendations and mitigate catastrophic forgetting. Our codes are shared on https://anonymous.4open.science/r/Critique-44F8.

The problem of maximizing the adoption of a product through viral marketing in social networks is of extreme importance and has been studied heavily through postulated network models. We present a novel data-driven formulation of the problem. We use Graph Neural Networks (GNNs) to model the adoption of products by utilizing both topological and attribute information. The resulting \emph{Dynamic Viral Marketing (DVM)} problem seeks to find the minimum budget and minimal set of dynamic topological and attribute changes in order to attain a specified adoption goal. We show that DVM is NP-Hard and is related to the existing influence maximization problem. Motivated by this connection, we develop the idea of Dynamic Gradient Influencing (DGI) that uses gradient ranking to find optimal perturbations and targets low-budget and high influence non-adopters in discrete steps. We use an efficient strategy for computing node budgets and develop the ``Meta-Influence'' heuristic for assessing a node’s downstream influence. We evaluate DGI against multiple baselines and demonstrate gains on average of 24\% on budget and 37\% on AUC on real-world attributed networks. Our code will be made publicly available.

Hypergraph neural networks (HyperGNNs) show promise in modeling online networks with high-order correlations. Despite notable progress, training these models on large-scale raw hypergraphs entails substantial computational and storage costs, thereby increasing the need of hypergraph size reduction. However, existing size reduction methods primarily capture pairwise association pattern within conventional graphs, making them challenging to adapt to hypergraphs with high-order correlations. To fill this gap, we introduce a novel hypergraph condensation framework, HG-Cond, designed to distill large-scale hypergraphs into compact, synthetic versions while maintaining comparable HyperGNN performance. Within this framework, we develop a Neural Hyperedge Linker to capture the high-order connectivity pattern through variational inference, achieving linear complexity with respect to the number of nodes. Moreover, We propose a multi-aspectual amelioration strategy including a Gradient-Parameter Synergistic Matching objective to holistically refine synthetic hypergraphs by coordinating improvements in node attributes, high-order connectivity, and label distributions. Extensive experiments demonstrate the efficacy of HG-Cond in hypergraph condensation, notably outperforming the original test accuracy on the 20News dataset while concurrently reducing the hypergraph size to a mere 5\% of its initial scale. Furthermore, the condensed hypergraphs demonstrate robust cross-architectural generalizability and potential for expediting neural architecture search. This research represents a significant advancement in hypergraph processing, providing a scalable approach for hypergraph-based learning in resource-limited environments.

The spread of rumors on social media, particularly during significant events like the US elections and the COVID-19 pandemic, poses a serious threat to social stability and public health. Current rumor detection methods primarily rely on propagation graphs to improve the model performance. However, the effectiveness of these methods is often compromised by noisy and irrelevant structures in the propagation process. To tackle this issue, techniques such as weight adjustment and data augmentation have been proposed. However, they depend heavily on rich original propagation structures, limiting their effectiveness in handling rumors that lack sufficient propagation information, especially in the early stages of dissemination. In this work, we introduce Key Propagation Graph Generator (KPG), a novel reinforcement learning-based framework, that generates contextually coherent and informative propagation patterns for events with insufficient topology information and identifies significant substructures in events with redundant and noisy propagation structures. KPG comprises two key components: the Candidate Response Generator (CRG) and the Ending Node Selector (ENS). CRG learns latent variable distributions from refined propagation patterns to eliminate noise and generate new candidates for ENS, while ENS identifies the most influential substructures in propagation graphs and provides training data for CRG. Furthermore, we develop an end-to-end framework that utilizes rewards derived from a pre-trained graph neural network to guide the training process. The resulting key propagation graphs are then employed in downstream rumor detection tasks. Extensive experiments conducted on four datasets demonstrate that KPG outperforms current state-of-the-art methods.