The proliferation of online hate speech poses a significant threat to the harmony of the web. While explicit hate is easily recognized through overt slurs, implicit hate speech is often conveyed through sarcasm, irony, stereotypes, or coded language---making it harder to detect. Existing hate speech detection models, which predominantly rely on surface-level linguistic cues, fail to generalize effectively across diverse stylistic variations. Moreover, hate speech spread on different platforms often targets distinct groups and adopts unique styles, potentially inducing spurious correlations between them and labels, further challenging current detection approaches. Motivated by these observations, we hypothesize that the generation of hate speech can be modeled as a causal graph involving key factors: contextual environment, creator motivation, target, and style. Guided by this graph, we propose CADET, a causal representation learning framework that disentangles hate speech into interpretable latent factors and then controls confounders, thereby isolating genuine hate intent from superficial linguistic cues. Furthermore, CADET allows counterfactual reasoning by intervening on style within the latent space, naturally guiding the model to robustly identify hate speech in varying forms. CADET demonstrates superior performance in comprehensive experiments, highlighting the potential of causal priors in advancing generalizable hate speech detection.
论文检索
输入标题、作者或关键词，从 7,537 篇学术成果中精准定位
Humans and large language models (LLMs) now co-produce and co-consume the web's shared knowledge archives. Such human-AI collective knowledge ecosystems contain feedback loops with both benefits (e.g., faster growth, easier learning) and systemic risks (e.g., quality dilution, skill reduction, model collapse). To understand such phenomena, we propose a minimal, interpretable dynamical model of the co-evolution of archive size, archive quality, model (LLM) skill, aggregate human skill, and query volume. The model captures two content inflows (human, LLM) controlled by a gate on LLM-content admissions, two learning pathways for humans (archive study vs. LLM assistance), and two LLM-training modalities (corpus-driven scaling vs. learning from human feedback). Through numerical experiments, we identify different growth regimes (e.g., healthy growth, inverted flow, inverted learning, oscillations), and show how platform and policy levers (gate strictness, LLM training, human learning pathways) shift the system across regime boundaries. Two domain configurations (PubMed, GitHub & Copilot) illustrate contrasting steady states under different growth rates and moderation norms. We also fit the model to Wikipedia's knowledge flow during pre-ChatGPT and post-ChatGPT eras separately. We find a rise in LLM additions with a concurrent decline in human inflow, consistent with a regime identified by the model. Our model and analysis yield actionable insights for sustainable growth of human-AI collective knowledge on the Web.
The predictive probability of the next token (P_token) in large language models (LLMs) is inextricably linked to the probability of relevance for the next piece of information, the purchase probability of the next product, and the execution probability of the next action-all of which fall under the scope of the task-level target distribution (P_task). While LLMs are known to generate samples that approximate real-world distributions, whether their fine-grained sampling probabilities faithfully align with task requirements remains an open question. Through controlled distribution-sampling simulations, we uncover a striking dichotomy in LLM behavior, distinguishing two model types: D-models (e.g. Qwen-2.5), whose P_token exhibits large step-to-step variability and poor alignment with P_task; and E-models (e.g. Mistral-Small), whose P_token is more stable and better aligned with P_task. We further evaluate these two model types in downstream tasks such as code generation and recommendation, revealing systematic trade-offs between diversity and stability that shape task outcomes. Finally, we analyze the internal properties of both model families to probe their underlying mechanisms. These findings offer foundational insights into the probabilistic sampling behavior of LLMs and provide practical guidance on when to favor D- versus E-models. For web-scale applications, including recommendation, search, and conversational agents, our results inform model selection and configuration to balance diversity with reliability under real-world uncertainty, providing a better level of interpretation.
Detecting harmful memes is crucial for safeguarding the integrity and harmony of online environments, yet existing detection methods are often resource-intensive, inflexible, and lacking explainability, limiting their applicability in assisting real-world web content moderation. We propose U-CoT+, a resource-efficient framework that prioritizes accessibility, flexibility and transparency in harmful meme detection by fully harnessing the capabilities of lightweight unimodal large language models (LLMs). Instead of directly prompting or fine-tuning large multimodal models (LMMs) as black-box classifiers, we avoid immediate reasoning over complex visual inputs but decouple meme content recognition from meme harmfulness analysis through a high-fidelity meme-to-text pipeline, which collaborates lightweight LMMs and LLMs to convert multimodal memes into natural language descriptions that preserve critical visual information, thus enabling text-only LLMs to ''see'' memes by ''reading''. Grounded in textual inputs, we further guide unimodal LLMs' reasoning under zero-shot Chain-of-Thoughts (CoT) prompting with targeted, interpretable, context-aware, and easily obtained human-crafted guidelines, thus providing accountable step-by-step rationales, while enabling flexible and efficient adaptation to diverse sociocultural criteria of harmfulness. Extensive experiments on seven benchmark datasets show that U-CoT+ achieves performance comparable to resource-intensive baselines, highlighting its effectiveness and potential as a scalable, explainable, and low-resource solution to support harmful meme detection.
The dominant paradigm for interacting with the Web has long relied on menu-driven navigation and complex graphical structures that can impose a high cognitive load on users. Recent advances in Large Language Models (LLMs) enable flexible NL understanding, yet purely conversational systems lack the stability and visual affordances required for complex tasks. This paper introduces Mirage, a framework that bridges these paradigms by making graphical user interface (GUI) components interpretable and orchestrable by LLMs, enabling on-the-fly construction of simplified UIs tailored to user requests. Developers declare application functions as reusable components with semantic descriptions and typed schemas; Mirage maps user goals, expressed in natural language, to the most relevant components and renders them as interactive interfaces. We implemented Mirage as a Next.js framework and validated it in a user study comparing the new paradigm to the navigation of a traditional website. Results show that Mirage significantly reduces task completion time, lowers cognitive workload, and achieves higher usability without compromising accuracy. Our approach also highlights new design directions toward supporting on-the-fly building of goal-driven interfaces.
Implicit hate speech often hides harmful intent behind innocuous wording, metaphors, or hostile tone, making it difficult for detectors that rely on surface cues. We observe that large language models (LLMs) frequently exhibit pseudo-reasoning that shows over-sensitivity to spurious cues while missing the latent semantic units that actually realize hateful intent. We call these units Latent Hate Components (LHCs) and argue they should be the anchors of inference. We propose DuPL, a Dual-Process argumentation framework that centers detection on LHCs. DuPL separates (i) Mining of LHCs via a high-recall Critical Miner and a Confusion Judge that early-exits on clear cases, from (ii) Deliberation of LHCs via component-wise argumentation and a final Integrative Decision. Across IHC, SBIC, and ToxiGen, DuPL consistently outperforms prompt-learning baselines, improving accuracy by +8.36, +4.93, and +3.78 percentage points and macro-F1 by +7.18, +5.00, and +3.99 percentage points, respectively. DuPL also lowers both false positive rate and false negative rate in most settings, indicating balanced mitigation of the two failure modes common in LLM detectors. By explicitly mining and deliberating over LHCs, DuPL turns opaque, uncontrolled reasoning into structured argumentation, yielding more accurate and interpretable decisions for web moderation.
Graph Neural Networks (GNNs) have achieved remarkable success across various graph-based tasks but remain highly sensitive to distribution shifts. In this work, we focus on a prevalent yet underexplored phenomenon in graph generalization, Minimal Shift Flip (MSF)—where test samples that slightly deviate from the training distribution are abruptly misclassified. To interpret this phenomenon, we revisit MSF through the lens of Sharpness-Aware Minimization (SAM), which characterizes the local stability and sharpness of the loss landscape while providing a theoretical foundation for modeling generalization error. To quantify loss sharpness, we introduce the concept of Local Robust Radius, measuring the smallest perturbation required to flip a prediction and establishing a theoretical link between local stability and generalization. Building on this perspective, we further observe a continual decrease in the robust radius during training, indicating weakened local stability and an increasingly sharp loss landscape that gives rise to MSF. To jointly solve the MSF phenomenon and the intractability of radius, we develop an energy-based formulation that is theoretically proven to be monotonically correlated with the robust radius, offering a tractable and principled objective for modeling flatness and stability. Building on these insights, we propose an energy-driven generative augmentation framework (E2A) that leverages energy-guided latent perturbations to generate pseudo-OOD samples and enhance model generalization. Extensive experiments across multiple benchmarks demonstrate that E2A consistently improves graph OOD generalization, outperforming state-of-the-art baselines. Code is available at https://github.com/anders1123/E2A
Federated graph learning (FGL) enables collaborative training without sharing raw graph data, but existing methods remain limited when clients perform different types of graph learning tasks. Recent advances in large language models (LLMs) provide new opportunities, where a promising solution is to project graph data into the LLM input space and leverage their generalization ability for prediction. However, directly applying LLMs to FGL raises two key challenges: (1) enhancing LLM interpretation of graphs with limited client data, and (2) obtaining a generalizable graph projector across heterogeneous client tasks. To address these challenges, we propose a LLM-enhanced federated graph learning framework FedLGS, which consists of two modules, i.e., geometry-aware graph projection (GGP) and shared subspace aggregation (SSA). GGP projects graph representations into complementary Riemannian manifolds with Ollivier-Ricci curvature priors, enabling geometry-aware and semantically enriched inputs for LLMs. SSA applies curvature-weighted whitening and aggregates updates in a shared subspace to achieve geometry-consistent generalization across heterogeneous tasks. Extensive experiments on six datasets demonstrate the effectiveness of FedLGS.
Text-attributed graphs (TAGs) enhance graph learning by integrating rich textual semantics and topological context for each node. While boosting expressiveness, they also expose new vulnerabilities in graph learning through text-based adversarial surfaces. Recent advances leverage diverse backbones, such as graph neural networks (GNNs) and pre-trained language models (PLMs), to capture both structural and textual information in TAGs. This diversity raises a key question: How can we design universal adversarial attacks that generalize across architectures to assess the security of TAG models? The challenge arises from the stark contrast in how different backbones—GNNs and PLMs—perceive and encode graph patterns, coupled with the fact that many PLMs are only accessible via APIs, limiting attacks to black-box settings. To address this, we propose BadGraph, a novel attack framework that deeply elicits large language models' (LLMs) understanding of general graph knowledge to jointly perturb both node topology and textual semantics. Specifically, we design a target influencer retrieval module that leverages graph priors to construct cross-modally aligned attack shortcuts, thereby enabling efficient LLM-based perturbation reasoning. Experiments show that BadGraph achieves universal and effective attacks across GNN- and LLM-based reasoners, with up to a 76.3% performance drop, while theoretical and empirical analyses confirm its stealthy yet interpretable nature.
Timely detection of anomalous interactions between different entities is crucial for the security and stability of Web-related systems, which are often modeled as streaming graphs. A major issue in existing anomaly detection methods is the effectiveness and the corresponding interpretability. In this paper, we introduce a novel evaluation metric, namely Interaction Willingness, to measure the propensity for entity interactions. Based on this metric, we design two efficient anomaly detection algorithms, AnoD and AnoC, tailored for real-time detection of two prevalent types of anomalous interaction behaviors (dense-type and counter-type), respectively. Notably, we adopt and extend the High-Order Count-Min Sketch for the implicit storage of streaming graphs, and ensure that both the space cost and the time cost of AnoD and AnoC for processing each new streaming edge remain constant and user-controllable, with estimation guarantees. Experimental evaluations on 6 real-world datasets demonstrate that the proposed evaluation metric and the associated algorithms achieve superior detection performance with clear interpretability for detecting anomalous interactions.
Signed Graph Neural Networks (SGNNs) have achieved outstanding performance in Link Sign Prediction (LSP), which involves predicting the existence and polarity of edges, by effectively modeling positive and negative interactions in signed graphs. However, their black-box nature raises transparency concerns, necessitating faithful explanations of model behavior to ensure trustworthiness and accountability. Existing eXplainable GNN (XGNN) methods, primarily designed for unsigned graphs, struggle to provide meaningful and human-understandable explanations for SGNN-based LSP, often generating disconnected subgraph explanations or neglecting the unique sign interactions. To address the gap, we propose SGExplainer, a novel method that leverages balanced paths, a concept rooted in signed graph theory, to provide clear and faithful explanations for LSP. SGExplainer employs a path-enforcing mask learning framework that ensures interpretable balanced path generation while maintaining explanation fidelity. Extensive experiments on real-world signed graphs demonstrate that SGExplainer consistently provides faithful and intuitive explanations for various SGNNs, outperforming state-of-the-art baselines in explanation quality, interpretability, and efficiency.
Graph Foundation Models (GFMs) have emerged as a frontier in graph learning, which are expected to deliver transferable representations across diverse tasks. However, GFMs remain constrained by in-memory bottlenecks: they attempt to encode knowledge into model parameters, which limits semantic capacity, introduces heavy lossy compression with conflicts, and entangles graph representation with the knowledge in ways that hinder efficient adaptation, undermining scalability and interpretability. In this work,we propose RAG-GFM, a Retrieval-Augmented Generation aided Graph Foundation Model that offloads knowledge from parameters and complements parameterized learning. To externalize graph knowledge, we build a dual-modal unified retrieval module, where a semantic store from prefix-structured text and a structural store from centrality-based motif. To preserve heterogeneous information, we design a dual-view alignment objective that contrasts both modalities to capture both content and relational patterns. To enable efficient downstream adaptation, we perform in-context augmentation to enrich supporting instances with retrieved texts and motifs as contextual evidence. Extensive experiments on five benchmark graph datasets demonstrate that RAG-GFM consistently outperforms 13 state-of-the-art baselines in both cross-domain node and graph classification, achieving superior effectiveness and efficiency.
Foundation models are at the forefront of artificial intelligence. A tokenizer, converting the raw input into discrete representations that the model can understand, plays an important role to the success of foundation models. Unlike the text tokenizer that is well studied in large language models, graph tokenizer is still at its early stage, facing the challenges of tackling the non-Euclidean structures and capturing the structural semantics. How to design a graph tokenizer for structural knowledge transfer? To this end, we propose a Riemannian Graph Tokenizer (RGT ) that bridges the structural knowledge and quantized representations to support cross-domain structural knowledge transfer. The connection is established by Riemannian geometry. Specifically, we first define the geometric vocabulary (trees, cycles and sequences), which captures fundamental structural patterns and reflects the intrinsic geometry of graph. Second, we construct a Riemannian quantizer with Riemannian Straight-Through Estimator to tokenise graph structures across multiple domains into discrete tokens. To ensure consistency and transferability across diverse geometric spaces, RGT further incorporates a geometry-aligned decoder that projects manifold-specific tokens into a unified tangent space. The theoretical analysis and geometric interpretations are provided to support the effectiveness of our proposed method. Extensive experiments across diverse datasets demonstrate that RGT significantly enhances structural knowledge transferability across graph domains.
Real-world pricing mechanisms are typically optimized using training data, a setting corresponding to the pricing query complexity problem in Mechanism Design. The previous work [11] studies the single-distribution case1, with tight bounds of ~Θ (ε-3 ) for a general distribution and ~Θ (ε-2 ) for either a regular or monotone-hazard-rate (MHR) distribution, where ε ∈ (0, 1) denotes the (additive) revenue loss of a learned uniform price relative to the Bayesian-optimal uniform price. This can be directly interpreted as ''the query complexity of the Uniform Pricing mechanism, in the single-distribution case''. Yet in the multi-distribution case, can the regularity and MHR conditions still lead to improvements over the tight bound ~Θ (ε-3) for general distributions? We answer this question in the negative, by establishing a (near-)matching lower bound Ømega(ε-3) for either two regular distributions or three MHR distributions. We also address the regret minimization problem and, in comparison with the folklore upper bound ~O(T2/3 ) for general distributions (see, e.g., [13]), establish a (near-)matching lower bound Ømega(T2/3 ) for either two regular distributions or three MHR distributions, via a black-box reduction. Again, this is in stark contrast to the tight bound ~Θ(T1/2 ) for a single regular or MHR distribution.
High dialogue engagement is a crucial indicator of an effective conversation. A reliable measure of engagement could help benchmark large language models, enhance the effectiveness of human-computer interactions, or improve personal communication skills. However, quantifying engagement is challenging, since it is subjective and lacks a ''gold standard''. This paper proposes PMIScore, an efficient unsupervised approach to quantify dialogue engagement. It uses pointwise mutual information (PMI), which is the probability of generating a response conditioning on the conversation history. Thus, PMIScore offers a clear interpretation of engagement. As directly computing PMI is intractable due to the complexity of dialogues, PMIScore learned it through a dual form of divergence. The algorithm includes generating positive and negative dialogue pairs, extracting embeddings by large language models (LLMs), and training a small neural network using a mutual information loss function. We validated PMIScore on both synthetic and real-world datasets. Our results demonstrate the effectiveness of PMIScore in PMI estimation and the reasonableness of the PMI metric itself.
Machine learning models for steel property prediction routinely report high-quality metrics with R² > 0.85, yet these results rely on random splits that allow similar grades in both train and test sets. We present SteelAgent, an interactive system that exposes a critical generalization gap: the same models drop from R² > 0.85 to R² = 0.11 on unseen steel families, revealing more than 7 times higher quality degradation. Similarly, conformal prediction coverage degrades from 91% to 38% under distribution shift induced by holding out substantial data sources. SteelAgent combines physics-informed features grounded in classical metallurgy and interpretable models with conformal uncertainty quantification, and an LLM orchestrator that coordinates six domain-specific tools. The system supports property prediction with specification compliance checking, competitive steel comparison, and cost-aware inverse alloy design over 3,741 heat treatment records spanning 1,234 grades. All predictions are traceable through explicit tool calls, ensuring that all physical quantities are computed, not generated. We made the code and data freely accessible to the community.
We present the Sparse ProtoPatient Demo, a publicly available interactive system for interpretable ICD-10 diagnosis prediction from clinical admission notes. The system is designed for clinicians in training, researchers, and educators exploring prototype-based diagnostic reasoning. The demo links predictions to learned prototypical patient representations and token-level evidence, allowing users to input custom text or select preset cases, inspect predicted ICD-10 codes, visualize label-wise saliency, retrieve supporting prototype notes, and compare alternative prototype cohorts. The demo provides a reproducible platform for interactive inspection of prototype-based clinical reasoning, enabling complementary opinion exploration, model auditing, and teaching of interpretable diagnosis prediction. The deployed model is trained on the publicly released CodiEsp corpus (1000 clinical notes, 955 ICD-10 labels) using a sparse multi-prototype architecture with five prototypes per label. We use the official machine-translated English CodiEsp-MT release to support English-language interaction. It achieves a macro-AUROC of 0.92 on a held-out test set and supports real-time interaction (300ms per query). The system is fully containerized for public research and educational use.
We present ElderMTL, a multi-task affect monitoring system designed for elderly care settings. The system simultaneously estimates Facial Action Units (FAUs), Valence-Arousal (VA) signals, and categorical emotions (FER) from video, capturing multiple layers of affective information. To improve sensitivity to subtle affective cues common in older adults, our approach incorporates age-conditioned physiological modeling, including baseline muscle adjustments and a dynamic AU co-activation graph. This enables the system to adapt to age-related changes in facial expression patterns, providing more reliable and interpretable emotion assessments. In a live demonstration, we showcase ElderMTL processing video streams, visualizing AU activations, affective state predictions, and interpretable insights that highlight age-specific affective dynamics. This work demonstrates that physiologically grounded, multi-task affective monitoring can provide meaningful, real-world support for elderly care.
Determining when to leave for the airport is a complex problem shaped by flight delay risk, traffic, weather, airport congestion, and passenger-specific constraints. Existing services rely on isolated delay estimates or simple travel time calculations, failing to capture real-time context. We propose ORBIT, a decision-support system that generates personalized leave-by recommendations by integrating predictive models with real-time operational signals. The system combines user input normalization, real-time data acquisition, Transformer-based delay prediction, and LLM-based reasoning to jointly account for statistical forecasts and dynamic factors like congestion, previous-leg propagation, weather, traffic, and airport processing times. Instead of a standalone delay estimate, ORBIT produces an actionable and interpretable departure plan. We implement ORBIT as an interactive system and demonstrate its applicability. A video demo is available at https://youtu.be/fZX42jceIM4.