WWW 2025 Papers

The Tor network, while offering anonymity through traffic routing across volunteer-operated nodes, remains vulnerable to attacks that aim to deanonymize users by correlating traffic patterns between colluded Entry and Exit nodes in circuits. This paper presents a novel approach for detecting anomalous circuits in the Tor network, and for the first time provides a more comprehensive identification of potential malicious accomplice nodes in Tor by taking roles of nodes in anomalous circuits into consideration. Our method strategically utilizes modified Middle nodes to capture traffic data, followed by a novel circuit classification based on traffic patterns to pinpoint concerned circuits. Two kinds of anomalies are identified: routing anomalies and usage anomalies, that respectively represent the anomalies with explicit or implicit violation of Tor's circuit construction guidelines. This leads to a successful revealing of totally 1,960 anomalous nodes in Tor. Furthermore, we apply clustering analysis with considering corresponding anomalous circuits and other key characteristics to the detected anomalous nodes, revealing potential hidden organizations behind these nodes that can threaten the network's security. Our findings highlight the necessity for the Tor project to adopt targeted mitigation strategies to enhance overall network security and privacy.

As web applications grow in popularity, developers are increasingly integrating deep learning (DL) models into these environments. Web-based DL frameworks (e.g., TensorFlow.js) are essential for building and deploying such applications. Ensuring the quality of these frameworks is critical for the reliability of DL systems. While extensive testing efforts have been made for native DL frameworks such as TensorFlow and PyTorch, web-based DL frameworks have not yet undergone systematic testing. A key challenge in this context is generating high-quality inputs that are both syntactically and semantically valid, as well as designing effective test oracles tailored to the unique constraints of web-specific environments. To address this gap, we introduce TensorJSFuzz, a novel method for testing web-based DL frameworks. To ensure input quality, TensorJSFuzz extracts constraints directly from the source code of framework APIs. By leveraging Large Language Models (e.g., ChatGPT) to understand the code and extract input constraints, TensorJSFuzz performs type-aware random generation coupled with dependency-aware refinement to create high-quality test inputs. These inputs are then subjected to differential testing across various backends, including CPU, TensorFlow, Wasm, and WebGL. Our experimental results show that TensorJSFuzz outperforms baseline methods in generating valid inputs and identifying bugs. In particular, TensorJSFuzz successfully detected 92 bugs, with 30 already confirmed or fixed by developers, demonstrating its effectiveness in improving the robustness of web-based DL frameworks.

Graph-level out-of-distribution (OOD) detection, which attempts to identify OOD graphs originated from an unknown distribution, is a vital building block for safety-critical applications in Web and society. Current approaches concentrate on how to learn better graph representations, but fail to provide any statistically guarantee on detection results, therefore impeding their deployments in the scenario where detection errors would result in serious consequences. To overcome this critical issue, we propose the Conformal Graph-level Out-of-distribution Detection (CGOD), extending the theory of conformal prediction to graph-level OOD detection with a rigorous control over the false positive rate. In CGOD, we develop a new aggregated non-conformity score function based on the proposed adaptive data augmentation. Through the guidance from two designed metrics, i.e., score consistency and representation diversity, our augmentation strategy can generate multiple non-conformity scores, and aggregating these generated non-conformity scores together is robust to the misleading information. Meanwhile, our score function can perceive the subsequent process of conformal inference, enabling the aggregated non-conformity score to be adaptive to different input graphs and deriving a more accurate score estimation. We conduct experiments on multiple real-world datasets with different empirical settings. Extensive results and model analyses demonstrate the superior performance of our approach over several competitive baselines.

The Web of Things (WoT) facilitates interoperability across web-based mobile and ubiquitous computing platforms and application domains, aiming to complement and preserve existing IoT standards and solutions. In this context, the multimodal federated learning (FL) paradigm has been introduced to enhance WoT by enabling the fusion of multi-source mobile sensing data while preserving privacy. However, a critical challenge in web-based mobile sensing systems employing multimodal FL is modality incompleteness, where certain modalities may be unavailable or partially captured, which can adversely impact the performance and reliability of these systems. Current multimodal FL frameworks typically train multiple unimodal FL subsystems or apply interpolation techniques on the node side to approximate missing modalities. However, these approaches overlook the shared latent feature space among incomplete modalities across different nodes and fail to discriminate against low quality nodes. To address this gap, we present FedMobile, a new knowledge contribution-aware multimodal FL framework designed for robust learning despite missing modalities. FedMobile prioritizes local-to-global knowledge transfer, leveraging cross-node multimodal feature information to reconstruct missing features. It also enhances system performance and resilience to modality heterogeneity through rigorous node contribution assessments and knowledge contribution-aware aggregation rules. Empirical evaluations on five widely recognized multimodal benchmark datasets demonstrate that FedMobile maintains robust learning even when up to 90% of modality information is missing or when data from two modalities is randomly missing, outperforming state-of-the-art baselines. Our datasets and code are available at the link.

Adversarial evasion attacks pose significant threats to graph learning, with lines of studies that have made progress in improving the robustness of Graph Neural Networks (GNNs) for real-world applications. However, existing works overly rely on priors of clean graphs or attacking strategies, which are often heuristic and not universally consistent. To achieve robust graph learning over different types of evasion attacks and diverse datasets, we investigate this non-trivial problem from a prior-free structure purification perspective. Specifically, we propose a novel **Diff**usion-based **S**tructure **P**urification framework named **DiffSP**, which creatively incorporates the graph diffusion model to learn intrinsic latent distributions of clean graphs and purify the perturbed structures by removing adversaries under the direction of the captured predictive patterns without relying on any pre-defined priors. DiffSP is divided into the forward diffusion process and the reverse denoising process, during which structure purification is achieved. To avoid valuable information loss during the forward process, we propose an LID-driven non-isotropic diffusion mechanism to selectively inject controllable noise anisotropically. To promote semantic alignment between the clean graph and the purified graph generated during the reverse process, we reduce the generation uncertainty by the proposed graph transfer entropy guided denoising mechanism. Extensive experiments on both graph and node classification tasks demonstrate the superior robustness of DiffSP against evasion attacks.

The "pre-training then fine-tuning" paradigm has advanced Graph Neural Networks (GNNs) by enabling the capture of general knowledge without task-specific labels. However, a significant objective gap between pre-training and downstream tasks limits their effectiveness. Recent graph prompting methods aim to bridge this gap by task reformulations and learnable prompts. Yet, they struggle with complex graphs like heterophily graphs—freezing the GNN encoder may diminish prompting effectiveness, and simple prompts fail to capture diverse hop-level distributions. This paper identifies two key challenges in adapting graph prompting methods for complex graphs: (i) adapting the model to new distributions in downstream tasks to mitigate pre-training and fine-tuning discrepancies from heterophily and (ii) customizing prompts for hop-specific node requirements. To overcome these challenges, we propose Distribution-aware Graph Prompt Tuning (DAGPrompT), which integrates a GLoRA module for optimizing the GNN encoder’s projection matrix and message-passing schema through low-rank adaptation. DAGPrompT also incorporates hop-specific prompts accounting for varying graph structures and distributions among hops. Evaluations on 10 datasets and 14 baselines demonstrate that DAGPrompT improves accuracy by up to 7.55\% in node and graph classification tasks, setting a new state-of-the-art while preserving efficiency. We provide our code and data via AnonymousGithub.

Large language models (LLMs) provide powerful foundations to perform fine-grained text re-ranking. However, they are often pro- hibitive in reality due to constraints on computation bandwidth. In this work, we propose a flexible architecture called Matroyshka Re-Ranker, which is designed to facilitate runtime customiza- tion of model layers and sequence lengths at each layer based on users’ configurations. Consequently, the LLM-based re-rankers can be made applicable across various real-world situations. The increased flexibility may come at the cost of precision loss. To address this problem, we introduce a suite of techniques to optimize the performance. First, we propose cascaded self-distillation, where each sub-architecture learns to preserve a precise re-ranking performance from its super components, whose predictions can be exploited as smooth and informative teacher signals. Second, we design a factorized compensation mechanism, where two col- laborative Low-Rank Adaptation modules, vertical and horizontal, are jointly employed to compensate for the precision loss resulted from arbitrary combinations of layer and sequence compression. We perform comprehensive experiments based on the passage and document retrieval datasets from MSMARCO, along with all public datasets from BEIR benchmark. In our experiments, Ma- tryoshka Re-Ranker substantially outperforms the existing meth- ods, while effectively preserving its superior performance across various forms of compression and different application scenarios. Our source code has been uploaded to this anonymous repository

The Web has shifted towards a mobile-first ecosystem with tools, frameworks, and forums explicitly discussing and catering for the mobile users, both mobile apps and mobile web-pages. Unfortunately much of the studies and designs are often based on analysis and findings from developed regions (e.g., N. America and Europe) or based on user-generated data (introducing bias). In this paper, we present one of the first studies to understand the interplay between hardware characteristics (e.g., cellular and mobile) on expected network and application level performance in Brazil (the largest developing region in S. America). We analyze more than 170 million measurement sessions collected from within the network of one of the largest Mobile Network Operators in Brazil. Our findings (1) illustrate limitations of existing crowdsourced measurements and inaccuracies in assumptions about adoption patterns and performance in the global south, (2) highlight the differences between recommendations made by standardization bodies and real world performance, (3) disclose a significant change pre- and post-pandemic, and (4) quantify the benefits of using both client side and network data for analysis.

In the era of data-centric AI, the focus of recommender systems has shifted from model-centric innovations to data-centric approaches. The success of modern AI models is built on large-scale datasets, but this also results in significant training costs. Dataset distillation has emerged as a key solution, condensing large datasets to accelerate model training while preserving model performance. However, condensing discrete and sequentially correlated user-item interactions, particularly with extensive item sets, presents considerable challenges. This paper introduces \textbf{TD3}, a novel \textbf{T}ucker \textbf{D}ecomposition based \textbf{D}ataset \textbf{D}istillation method within a meta-learning framework, designed for sequential recommendation. TD3 distills a fully expressive \emph{synthetic sequence summary} from original data. To efficiently reduce computational complexity and extract refined latent patterns, Tucker decomposition decouples the summary into four factors: \emph{synthetic user latent factor}, \emph{temporal dynamics latent factor}, \emph{shared item latent factor}, and a \emph{relation core} that models their interconnections. Additionally, a surrogate objective in bi-level optimization is proposed to align feature spaces extracted from models trained on both original data and synthetic sequence summary beyond the na\"ive performance matching approach. In the \emph{inner-loop}, an augmentation technique allows the learner to closely fit the synthetic summary, ensuring an accurate update of it in the \emph{outer-loop}. To accelerate the optimization process and address long dependencies, RaT-BPTT is employed for bi-level optimization. Experiments and analyses on multiple public datasets have confirmed the superiority and cross-architecture generalizability of the proposed designs. Codes are released at \textcolor{blue}{\url{https://anonymous.4open.science/r/TD3}}.

We show that the market for non-fungible tokens (NFTs), much like the luxury fashion market, exhibits conspicuous consumption dynamics: an NFT's value depends substantially on its social meaning as a signal of wealth, taste, and community affiliation. More specifically, we introduce a novel dataset of NFT transaction data combined with embeddings of the corresponding NFT images computed using an off-the-shelf vision transformer architecture. We use our dataset to identify evidence for two phenomena that prior work has identified as the primary determinants of conspicuous consumption: the \emph{bandwagon effect} and the \emph{snob effect}. For each determinant, we identify characteristics of the NFTs themselves and of the communities surrounding them that drive the effect.

An important problem on social information sites is the recovery of ground truth from individual reports when the experts are in the minority. The wisdom of the crowd, i.e. the collective opinion of a group of individuals fails in such a scenario. However, the surprisingly popular (SP) algorithm~\cite{prelec2017solution} can recover the ground truth even when the experts are in the minority, by asking the individuals to report additional prediction reports -- their beliefs about the reports of others. Several recent works have extended the surprisingly popular algorithm to an equivalent voting rule (SP-voting) to recover the ground truth ranking over a set of $m$ alternatives. However, we are yet to fully understand when SP-voting can recover the ground truth ranking, and if so, how many samples (votes and predictions) it needs. We answer this question by proposing two rank-order models and analyzing the sample complexity of SP-voting under these models. In particular, we propose concentric mixtures of Mallows and Plackett-Luce models with $G (\ge 2)$ groups. Our models generalize previously proposed concentric mixtures of Mallows models with $2$ groups, and we highlight the importance of $G > 2$ groups by identifying three distinct groups (expert, intermediate, and non-expert) from existing datasets. Next, we provide conditions on the parameters of the underlying models so that SP-voting can recover ground-truth rankings with high probability, and also derive sample complexities under the same. We complement the theoretical results by evaluating SP-voting on simulated and real datasets.

Temporal knowledge graph completion (TKGC) seeks to develop more comprehensive knowledge representations by addressing missing relationships and entities within temporal knowledge graphs (TKGs), thereby enhancing reasoning and predictive capabilities in downstream tasks. Nonetheless, real-world knowledge—such as the progression of social network interactions and the unfolding of news events—is inherently dynamic, resulting in substantial sparsity issues in TKGs that profoundly impair the performance of TKGC models. To overcome this challenge, we introduce the Adaptive Neighborhood Enhancement Layer (ANEL), a novel module that can be effortlessly integrated into existing TKGC models to substantially elevate the representation quality of sparse entities. ANEL first derives initial entity embeddings through a base model and then uncovers concealed semantic relationships between entities via a latent relation module, enriching the explicit relationships within the knowledge graph. Furthermore, ANEL incorporates an adaptive latent information adjustment component, which dynamically calibrates the influence of latent information based on the entity's relational structure: entities with fewer connections derive greater benefit from latent information, while entities with denser connections become less dependent on latent augmentation, ensuring precise and resilient representations. We conducted comprehensive experiments on four prominent benchmark datasets, and the results underscore the effectiveness and superiority of ANEL in TKGC tasks. The code is available at: https://anonymous.4open.science/r/ANEL-177F.

The success of Graph Neural Networks (GNNs) in graph classification has heightened interest in explainable GNNs, particularly through graph rationalization. This method aims to enhance GNNs explainability by identifying subgraph structures (i.e., rationales) that support model predictions. However, existing methods often rely on centralized datasets, posing challenges in scenarios where data privacy is crucial, such as in molecular property prediction. Federated Learning (FL) offers a solution by enabling collaborative model training without sharing raw data. In this context, Federated Graph Rationalization emerges as a promising research direction. However, in each client, the rationalization methods often rely on client-specific shortcuts to compose rationales and make task predictions. Data heterogeneity, characterized by non-IID data across clients, exacerbates this problem, leading to poor prediction performance. To address these challenges, we propose the Environment-aware Data Augmentation (EaDA) method for Federated Graph Rationalization. EaDA comprises two main components: the Environment-aware Rationale Extraction (ERE) module and the Local-Global Alignment (LGA) module. The ERE module employs prototype learning to infer and share abstract environment information across clients, which are then aggregated to form a global environment. This information is used to generate counterfactual samples for local clients, enhancing the robustness of task predictions. The LGA module uses contrastive learning methods to align local and global rationale representations, mitigating performance degradation due to data heterogeneity. Comprehensive experiments on benchmark datasets demonstrate the effectiveness of our approaches.

Diffusion models (DMs) have emerged as the new state-of-the-art family of deep generative models. To gain deeper insights into the limitations of diffusion models in recommender systems, we investigate the fundamental structural disparities between images and items. Consequently, items often exhibit distinct anisotropic and directional structures that are less prevalent in images. However, the traditional forward diffusion process continuously adds isotropic Gaussian noise, causing anisotropic signals to degrade into noise, which impairs the semantically meaningful representations in recommender systems. Inspired by the advancements in hyperbolic spaces, we propose a novel \textbf{H}yperbolic \textbf{D}iffusion \textbf{R}ecommender \textbf{M}odel (named HDRM). Unlike existing directional diffusion methods based on Euclidean space, the intrinsic non-Euclidean structure of hyperbolic space makes it particularly well-adapted for handling anisotropic diffusion processes. In particular, we begin by constructing a geometrically latent space grounded in hyperbolic geometry, incorporating interpretability measures to define the latent anisotropic diffusion processes. Subsequently, we propose a novel hyperbolic latent diffusion process specifically tailored for users and items. Drawing upon the natural geometric attributes of hyperbolic spaces, we restrict both radial and angular components to facilitate directional diffusion propagation, thereby ensuring the preservation of the original topological structure in user-item interaction graphs. Extensive experiments on three benchmark datasets demonstrate the effectiveness of HDRM. Our code is available at \url{https://anonymous.4open.science/status/HDRM-ECFA}.

Online continual learning (CL) is becoming a mainstream paradigm to learn incrementally from task streams without forgetting previously learned knowledge. However, current online CL primarily focuses on the learning performance, such as avoiding catastrophic forgetting, neglecting the critical demands of real-time inference. As a result, the performance of real-time inference in online CL degrades significantly due to frequent data distribution variations and time-consuming incremental model adaptation. In this work, we propose ELITE, an online CL framework with device-cloud collaboration, to realize on-device real-time inference on time-varying task streams with performance guarantee. To realize on-device real-time inference in online CL, ELITE features a new design of the model zoo comprising various pre-trained models with the assistance of the cloud, and proposes a task-oriented on-device model selection to quickly retrieve the best-fit models instead of performing time-consuming model retraining. To prevent performance degradation on new tasks not available in the cloud, we introduces a latency-aware on-device model fine-tuning strategy to adapt to new tasks with accuracy-latency trade-off, and dynamically updates the model zoo in the cloud to enhance ELITE. Extensive evaluations on five real-world datasets have been conducted, and the results demonstrate that ELITE consistently outperforms the state-of-art solutions, improving the accuracy by 16.3\% on average and reducing the response latency by up to 1.98 times.

In online advertising, uncertainty calibration aims to adjust a ranking model's probability predictions to better approximate the true likelihood of an event, e.g., a click or a conversion. However, existing calibration approaches may lack the ability to effectively model complex nonlinear relations, consider context features, and achieve balanced performance across different data subsets. To tackle these challenges, we introduce a novel model called Monotonic Calibration Networks, featuring three key designs: a monotonic calibration function (MCF), an order-preserving regularizer, and a field-balance regularizer. The nonlinear MCF is capable of naturally modeling and universally approximating the intricate relations between uncalibrated predictions and the posterior probabilities, thus being much more expressive than existing methods. MCF can also integrate context features using a flexible model architecture, thereby achieving context awareness. The order-preserving and field-balance regularizers promote the monotonic relationship between adjacent bins and the balanced calibration performance on data subsets, respectively. Experimental results on both public and industrial datasets demonstrate the superior performance of our method in generating well-calibrated probability predictions.

Post-quantum cryptography (PQC) aims to develop quantum-safe algorithms against attacks by a quantum computer. As quantum-safe algorithms require much larger keys in their operation compared to the current RSA/ECC practice, the networking latency significantly increases when executing the protocols with sending such large keys. This problem gets more challenging in the era of Web-of-Things (WoTs) with low-memory devices. To tackle the problem, we propose ExpressPQDelivery, which is, to the best of our knowledge, the first immediately deployable protocol to efficiently transport large keys. It leverages the DNS infrastructure, as DNS is close to clients, guaranteeing express key delivery with a short round-trip time (RTT). We split a large PQ key along with a server's signature and feed them into several DNS records. To show the feasibility of ExpressPQDelivery, we instantiate it with TLS 1.3 and demonstrate that it reduces 27\% of network latency between a server and a client on average compared to the standard TLS 1.3. We deploy ExpressPQDelivery on a low-capability board with 256 KB RAM, showing a significant high gain (34\%).

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.