Large Language Models (LLMs) have shown strong potential as conversational agents. Yet, their effectiveness remains limited by deficiencies in robust long-term memory—particularly in complex, long-term Web-based services such as online emotional support. However, existing long-term dialogue benchmarks primarily focus on static and explicit fact retrieval, failing to evaluate agents in these critical scenarios where user information is dispersed, implicit, and continuously evolving. To address this gap, we introduce ES-MemEval, a comprehensive benchmark that systematically evaluates five core memory capabilities—information extraction, temporal reasoning, conflict detection, abstention, and user modeling—in long-term emotional support scenarios, covering question answering, summarization, and dialogue generation tasks. To support the benchmark, we also propose EvoEmo, the first multi-session dataset for personalized long-term emotional support scenarios, capturing fragmented, implicit user disclosures and evolving user states. Extensive experiments on open-source long-context, commercial, and retrieval-augmented (RAG) LLMs reveal that explicit long-term memory is essential to reduce hallucinations and enable effective personalization. At the same time, RAG enhances factual consistency but struggles with temporal dynamics and evolving user states. These findings highlight both the potential and limitations of current paradigms, encouraging the development of more robust memory–retrieval integration in long-term personalized dialogue systems.
论文检索
输入标题、作者或关键词，从 12,319 篇学术成果中精准定位
The rapid growth of Web, mobile, and Web of Things (WoT) applications increasingly relies on Federated Learning (FL) to enable privacy-preserving intelligence across distributed and heterogeneous devices. However, the global model in FL often suffers from performance degradation due to global knowledge forgetting caused by highly non-IID client data. While client-side Knowledge Distillation (KD) has emerged as a promising paradigm for transferring global knowledge to local models, existing approaches inadvertently distort global knowledge through local data priors, fundamentally limiting their effectiveness in real-world Web and mobile environments. We further find that this knowledge forgetting is asymmetric, with locally under-represented classes suffering significantly more severe performance degradation than over-represented ones. While the conventional KD inherently couples these two distinct kinds of knowledge, thereby hindering the targeted preservation of critical global knowledge for under-represented classes. To address these challenges, we propose Federated Adaptive Knowledge Distillation (FedAKD), a novel framework that adaptively calibrates and decouples global knowledge according to the local data distribution, enabling precise and flexible distillation across heterogeneous clients. Extensive experiments on benchmark datasets and diverse data heterogeneity settings demonstrate that FedAKD achieves state-of-the-art performance, offering a scalable and sustainable solution for Web, mobile, and WoT intelligence deployment.
Large Language Models (LLMs) increasingly underpin intelligent web applications, from chatbots to search and recommendation, where efficient specialization is essential. Low-Rank Adaptation (LoRA) enables such adaptation with minimal overhead, while federated LoRA allows web service providers to fine-tune shared models without data sharing. However, in privacy-sensitive deployments, clients inject varying levels of differential privacy (DP) noise, creating privacy heterogeneity that misaligns individual incentives and global performance. In this paper, we propose WinFLoRA, a privacy-heterogeneous federated LoRA that utilizes aggregation weights as incentives with noise awareness. Specifically, the noises from clients are estimated based on the uploaded LoRA adapters. A larger weight indicates greater influence on the global model and better downstream task performance, rewarding lower-noise contributions. By up-weighting low-noise updates, WinFLoRA improves global accuracy while accommodating clients' heterogeneous privacy requirements. Consequently, WinFLoRA aligns heterogeneous client utility in terms of privacy and downstream performance with global model objectives without third-party involvement. Extensive evaluations demonstrate that across multiple LLMs and datasets, WinFLoRA achieves up to 52.58% higher global accuracy and up to 2.56× client utility than state-of-the-art benchmarks. Source code is publicly available at https://github.com/koums24/WinFLoRA.git.
Trace analysis is essential for understanding system behaviors, detecting anomalies, and diagnosing faults in complex microservice-based web applications. Existing trace analysis approaches face several challenges in industrial microservice-based systems, including high manual overhead, limited functionality, unfriendly interaction mechanisms, and difficulties in deployment and integration. The strong capabilities of large language models (LLMs) in natural language understanding, reasoning, and multi-task generalization provide new opportunities for a more intelligent and flexible trace analysis approach. However, the trace analysis capabilities of LLMs remain underexplored and underdeveloped. To bridge this gap, we conduct the first comprehensive evaluation on the trace analysis capabilities of LLMs. In particular, we construct the first instruction&response benchmark dataset for trace analysis, named TraceBench. It involves a wide range of trace analysis tasks, allowing us to systematically evaluate the capabilities of LLMs in this area. Experimental results show that LLMs have potential in handling trace analysis tasks, but there leaves room for improvement. To this end, we propose TraceLLM, an approach that significantly enhances the capabilities of LLMs via fine-tuning, outperforming the open-source LLMs by 34.77% on average in terms of accuracy, and outperforming the closed-source model by 21.66% in the best case. The generalization and robustness of TraceLLM are also confirmed in our experiments. To the best of our knowledge, TraceLLM is the first LLM which is specialized for handling various types of trace analysis tasks. This work provides a foundation for future research to further explore the trace analysis capabilities of LLMs.
Depressive disorders represent a major global public health challenge. As an increasing number of individuals share their emotional experiences and concerns on social media, researchers have shown growing interest in leveraging such data for early depression screening. However, most existing methods rely on a fixed model and a singular reasoning paradigm, which constrains their adaptability to depression detection. The limited availability of mental health-related data and variability in training data distributions across different LLMs hinder their consistent and comprehensive understanding of diverse psychological symptoms. In this paper, we propose AdaDepression, a framework that enables explainable depression screening through a two-hop retrieval algorithm to identify symptom-relevant posts and a two-stage adaptive routing mechanism for selecting appropriate reasoning strategies and LLMs. Specifically, we first collect representative posts from the training dataset to capture the real-world symptom expressions, and then utilize these posts to retrieve symptom-relevant posts from the user's posting history. Subsequently, we employ the Mixture of Routers (MoR), which integrates the Mixture of Experts (MoE) into the routing mechanism to select the optimal reasoning strategies and LLMs in a cascaded manner. Finally, we complete the standardized psychological questionnaire using the selected LLMs and reasoning strategies. Experimental results on the Reddit-based benchmarks demonstrate the effectiveness of the proposed method, outperforming existing studies on various metrics. Our code is released at https://github.com/MindIntLab-HFUT/AdaDepression.
Dynamic link prediction plays a crucial role in diverse applications including social network analysis, communication forecasting, and financial modeling. While recent Transformer-based approaches have demonstrated promising results in temporal graph learning, their performance remains limited when capturing complex multi-scale temporal dynamics. In this paper, we propose TFWaveFormer, a novel Transformer architecture that integrates temporal-frequency analysis with multi-resolution wavelet decomposition to enhance dynamic link prediction. Our framework comprises three key components: (i) a temporal-frequency coordination mechanism that jointly models temporal and spectral representations, (ii) a learnable multi-resolution wavelet decomposition module that adaptively extracts multi-scale temporal patterns through parallel convolutions, replacing traditional iterative wavelet transforms, and (iii) a hybrid Transformer module that effectively fuses local wavelet features with global temporal dependencies. Extensive experiments on benchmark datasets demonstrate that TFWaveFormer achieves state-of-the-art performance, outperforming existing Transformer-based and hybrid models by significant margins across multiple metrics. The superior performance of TFWaveFormer validates the effectiveness of combining temporal-frequency analysis with wavelet decomposition in capturing complex temporal dynamics for dynamic link prediction tasks. The code is available at https://github.com/SEUFHTong/TFWaveFormer.
Opinion propagation research primarily focuses on phenomenon prediction rather than mechanism understanding, lacking interpretable frameworks to reveal underlying propagation dynamics. This limitation stems from two sources: existing methods employ end-to-end paradigms where parameters lack physical meanings, while available datasets suffer from incomplete hierarchical structures, coarse sentiment annotations, and limited domain coverage. To address these limitations, we introduce VISTA, a multi-dimensional opinion propagation dataset providing complete hierarchical structures, fine-grained emotional annotations, and cross-domain coverage. Based on this dataset, we propose an interpretable modeling framework integrating high-dimensional Hawkes processes with graph neural networks, enabling parametric expression of propagation mechanisms through event space constructed from emotional and reply level combinations. Through interpretable parameter analysis, we reveal three mechanistic patterns: differential emotional propagation strength, asymmetric hierarchical excitation, and temporal memory effects. Our framework establishes quantitative foundations for understanding opinion propagation dynamics, achieving best performance in sentiment prediction and structural consistency tasks while providing the first benchmark for multi-dimensional propagation mechanism analysis.
Graph neural networks (GNNs) have become the cornerstone of graph representation learning in real-world applications, especially in social networks such as Facebook and Twitter, where user interactions naturally form graph-structured data. To safeguard sensitive information, including user attributes and personal connections, locally private graph learning protocols employ local differential privacy (LDP) to provide rigorous user-level guarantees. Despite these protections, we show that such protocols remain highly vulnerable to data poisoning attacks. Our approach targets the full-privacy setting, where both node features and edges are LDP-protected, and executes coordinated manipulations that significantly degrade utility, such as node classification accuracy, across multiple social network benchmarks. To counter these threats, we propose CureNet, a defense framework with four key components: (1) local data perturbation for privacy, (2) trimmed screening to filter abnormal submissions, (3) privacy-aware fake node pruning to remove sophisticated adversaries, and (4) a utility enhancement module to recover graph learning performance under privacy constraints. Extensive experiments on four real-world social network datasets—LastFM, Twitch, Github, and Facebook—demonstrate both the severe impact of our attack and the effectiveness of our defense. This study underscores the critical need to secure privacy-preserving graph learning against sophisticated poisoning threats.
Retrieval-augmented generation (RAG) has improved large language models (LLMs) on knowledge-intensive tasks, yet most systems assume static facts and struggle when answers depend on serialized and dynamic data, like time--e.g., ordering events, aligning facts to valid intervals, or planning actions under evolving conditions. This paper presents Knowledge-Graph Iterative Retrieval-Augmented Generation (KG-iRAG), a framework specialized for temporal reasoning. KG-iRAG couples a time-aware planner with a knowledge graph (KG) to iteratively fetch and compose evidence along a temporal axis. Concretely, it (i) represents events and facts with explicit timestamps and validity intervals; (ii) propagates temporal constraints through iterative retrieval using operators; and (iii) verifies temporal consistency while refining intermediate hypotheses, enabling step-by-step deduction for queries that mix knowledge retrieval with inference. Across public temporal QA benchmarks, KG-iRAG consistently improves accuracy and calibration over strong RAG baselines while reducing unnecessary retrieval through targeted, constraint-guided steps. To stress-test real-time decision queries, three application-oriented datasets (weatherQA-Irish, ~ weatherQA-Sydney, and~ trafficQA-TFNSW) are additionally constructed and tested alongside existing temporal benchmarks. The results demonstrate that injecting temporal structure into KG-driven RAG yields robust gains on multi-step, time-dependent queries, advancing the state of temporal reasoning with LLMs.
Taxonomy completion is the task of integrating new concepts into an existing taxonomy by determining the appropriate hypernym--hyponym relations. Existing approaches often struggle with the inherent imbalance between leaf and non-leaf edges, which induces bias in representation learning. In this paper, we propose BLEND: Balanced and Leaf-Enhanced Dual Fine-Tuning for Taxonomy Completion, a novel framework designed to mitigate this inductive bias. Our method employs independent fine-tuning of two lightweight large language models (LLMs): one optimized with a leaf-focused objective and the other trained with a balanced focused strategy. To further enhance structural understanding, we apply contrastive learning over structure-encoded paths and introduce a combined loss function, enabling more robust representation of hierarchical relations. Extensive experiments on three real-world benchmark datasets demonstrate that BLEND achieves up to 9.32% improvement in recall or hit metrics compared to state-of-the-art approaches. Moreover, BLEND delivers efficient inference while outperforming the latest baseline COMI, highlighting its effectiveness for taxonomy completion tasks.
Transparent and standardized documentation is essential for building trustworthy generative AI (GAI) systems. However, current automated model and data card generation methods still face three key challenges: (i) Static templates. Most systems rely on fixed query templates that cannot adapt to diverse paper structures or evolving documentation requirements. (ii) Information scarcity. Web-scale repositories such as Hugging Face often provide incomplete or inconsistent metadata, resulting in missing or noisy information. (iii) Lack of benchmarks. The absence of standardized datasets and evaluation protocols prevents fair and reproducible assessment of documentation quality. To address these challenges, we propose AdaQE-CG, an Adaptive Query Expansion for Card Generation framework that integrates dynamic information extraction with cross-card knowledge transfer. The Intra-Paper Extraction via Context-Aware Query Expansion (IPE-QE) module iteratively refines extraction queries to capture richer and more complete information from scientific papers and repositories. The Inter-Card Completion using the MetaGAI Pool (ICC-MP) module enriches missing fields by transferring semantically relevant content from similar cards within a curated dataset. In addition, we construct MetaGAI-Bench, the first large-scale, expert-annotated benchmark for evaluating GAI documentation. Comprehensive experiments across five quality dimensions demonstrate that AdaQE-CG significantly outperforms existing approaches, surpasses human-authored data cards, and approaches human-level quality for model cards. Code, prompts, and data are publicly available at: https://github.com/haoxuan-unt2024/AdaQE-CG.
Taxonomies form the backbone of structured knowledge representation across diverse domains, enabling applications such as e-commerce and semantic search. Yet, manual taxonomy expansion is labor-intensive and slow. Existing methods rely on point-based vector embeddings, which model symmetric similarity and thus struggle with the asymmetric relationships that are fundamental to taxonomies. Box embeddings offer a promising alternative by enabling containment and disjointness, but they face key issues: (i) unstable gradients at the intersection boundaries, (ii) no notion of semantic uncertainty, and (iii) limited capacity to represent polysemy or ambiguity. We address these shortcomings with TaxoBell, a Gaussian box embedding framework that translates between box geometries and multivariate Gaussian distributions, where means encode semantic location and covariances encode uncertainty. Energy-based optimization yields stable optimization, robust modeling of ambiguous concepts, and interpretable hierarchical reasoning. Extensive experiments on five benchmark datasets demonstrate that TaxoBell significantly outperforms eight state-of-the-art taxonomy expansion baselines by 19% in MRR and around 25% in Recall@k. We further demonstrate the advantages and pitfalls of TaxoBell with error analysis and ablation studies.
The growing scale of knowledge graphs demands scalable systems for their subsequent processing. However, accurate benchmarking requires large knowledge graphs. While data-driven synthetic generators based on versioned datasets are promising to generate large realistic graphs, current approaches generate the graph at a triple level without considering higher-order structures. This work introduces SimplexKG, a simplex-based synthetic knowledge graph generator. Our approach analyzes d-dimensional simplices within input knowledge graphs and leverages the identified simplicial networks to generate a synthetic graph of arbitrary size. We explore whether leveraging higher-dimensional structures enhances the realism of synthetic graphs by evaluating the structure and the utility of the generated graphs. Our approach consistently outperforms 2 baseline generators and 6 variants of the state-of-the-art generator LEMMING in structural fidelity and triple store benchmarking scenarios across 3 datasets. Specifically, compared to the second-best approach, our graphs achieve a structuredness value up to 26.62 % closer to the target graph, while reducing the query throughput error by up to 6.59 % across storage solutions.
Analogical reasoning over Multimodal Knowledge Graphs (MMKGs) couples abductive relation induction with inductive tail completion. However, existing approaches rely on static fusion mechanisms that overlook the inherent asymmetry of modal relevance: while visual cues elucidate concrete entities, they are often noisy or irrelevant for abstract concepts, where text and graph structure provide decisive signals. Furthermore, prior methods fail to enforce consistency between induced relations and the modality patterns implied by the analogical context. To bridge this gap, we introduce RMAR, a Relation-aware Multimodal Analogical Reasoning framework with two complementary paths. An explicit path estimates modality fingerprints to score compatibility during relation induction and guide fusion during tail completion. An implicit path employs adaptive gating to blend structural, textual, and visual signals conditioned on the specific query context. To address the limitations of current benchmarks, which overrepresent concrete entities, we release MCNetAnalogy, and its companion graph, MCNetKG, a rigorous dataset enriched with abstract concepts and actions. RMAR is backbone-agnostic and works with multimodal knowledge graph embedding (MKGE) and transformer-based (MPT) pipelines. Extensive experiments demonstrate that RMAR delivers consistent gains across both embedding-based and transformer-based backbones, achieving a 29% relative improvement on MCNetAnalogy. Ablation studies confirm that RMAR's relation-aware modulation is particularly effective when modal evidence is weak or ambiguous.
Multi-modal entity alignment (MMEA) aims to identify equivalent entities across heterogeneous multi-modal knowledge graphs (MMKGs), which play a crucial role in organizing and integrating web knowledge from diverse modalities. Although prior studies have made progress by multi-modal features fusion, three inherent limitations remain unresolved. First, instance-level feature fusion is misaligned with the pair-wise task format of MMEA. Second, joint representations often overlook modality-specific characteristics, resulting in insufficient alignment. Third, most existing methods rely solely on global features of modality. This may lead to the misalignment of entities that are similar yet distinct. To address above issues, we propose DMEA, a new decision-fusion-based framework. Specifically, we first design a multi-modal knowledge encoding module to extract both global and local features for different modalities and then introduce a multi-grained alignment module, which consists of two components: intra-modal alignment and cross-modal alignment. The former computes alignment scores between the global and local features of entity pairs within the same modality, while the latter leverages the complementarity across modalities to compute cross-modal alignment scores. Each score is regarded as an independent decision, and the final alignment judgment is made by integrating all decisions. Finally, we incorporate an intra-modal contrastive loss to obtain more discriminative embedding representations. DMEA achieves improvements of 13.2% and 14.8% in hit@1 over the state-of-the-art models on two benchmark datasets, FB15K-DB15K and FB15K-YAGO15K, respectively, validating the superiority of our framework.
Large language models (LLMs) are increasingly adopted as the backbone of recommender systems. However, user–item interactions in real-world scenarios are non-stationary, making preference drift over time inevitable. Existing model update strategies mainly rely on global fine-tuning or pointwise editing, but they face two fundamental challenges: (i) imbalanced update granularity, where global updates perturb behaviors unrelated to the target while pointwise edits fail to capture broader preference shifts; (ii) unstable incremental updates, where repeated edits interfere with prior adaptations, leading to catastrophic forgetting and inconsistent recommendations. To address these issues, we propose Region-Aware Incremental Editing (RAIE), a plug-in framework that freezes the backbone model and performs region-level updates. RAIE first constructs semantically coherent preference regions via spherical k-means in the representation space. It then assigns incoming sequences to regions via confidence-aware gating and performs three localized edit operations-Update, Expand, and Add-to dynamically revise the affected region. Each region is equipped with a dedicated Low-Rank Adaptation (LoRA) module, which is trained only on the region's updated data. During inference, RAIE routes each user sequence to its corresponding region and activates the region-specific adapter for prediction. Experiments on two benchmark datasets under a time-sliced protocol that segments data into Set-up (S), Finetune (F), and Test (T) show that RAIE significantly outperforms state-of-the-art baselines while effectively mitigating forgetting. These results demonstrate that region-aware editing offers an accurate and scalable mechanism for continual adaptation in dynamic recommendation scenarios.
Effective representations of complex web semantic graphs are essential for various web applications, including link prediction, recommendation systems, and social network analysis. However, existing methods assume that multi-interactions (or multi-relationships) between two connected nodes are independent, while these relationships inherently exhibit characteristics of mutual promotion or mutual inhibition. Moreover, these semantic characteristics across different relationships cannot be easily captured by a simple linear combination. To tackle this challenge, we propose an Adaptive Multi-Interaction (AMI) web semantic graph representation method. Specifically, AMI consists of three modules, including a multi-interaction aggregation module, a global pattern aggregation module, and an adaptive relation-specific decoder module. Firstly, we construct a learnable multi-interaction behavior pattern matrix that captures the mutual promotion and mutual inhibition effects between two connected nodes. Secondly, the global pattern aggregation module is designed to efficiently capture global homogeneous interaction patterns through graph convolution networks. Finally, the adaptive relation-specific decoder module employs a hybrid scoring strategy to adaptively decode node embeddings based on their distinct relationships. Extensive experiments on benchmark web datasets for link prediction tasks demonstrate that AMI outperforms state-of-the-art baselines. Our codes are available at https://github.com/AI-stronger123/AMI.
Large Language Models (LLMs) have recently emerged as a powerful paradigm for Knowledge Graph Completion (KGC), offering strong reasoning and generalization capabilities beyond traditional embedding-based approaches. However, existing LLM-based methods often struggle to fully exploit structured semantic representations, as the continuous embedding space of pretrained KG models is fundamentally misaligned with the discrete token space of LLMs. This discrepancy hinders effective semantic transfer and limits their performance. To address this challenge, we propose ReaLM, a novel and effective framework that bridges the gap between KG embeddings and LLM tokenization through the mechanism of residual vector quantization. ReaLM discretizes pretrained KG embeddings into compact code sequences and integrates them as learnable tokens within the LLM vocabulary, enabling seamless fusion of symbolic and contextual knowledge. Furthermore, we incorporate ontology-guided class constraints to enforce semantic consistency, refining entity predictions based on class-level compatibility. Extensive experiments on two widely used benchmark datasets demonstrate that ReaLM achieves state-of-the-art performance, confirming its effectiveness in aligning structured knowledge with large-scale language models. The implementation is publicly available at https://github.com/xiumu-gg/ReaLM.
Semantic graphs are fundamental to the Web, enabling applications such as semantic search, recommendation, and knowledge-intensive reasoning. In decentralized Web environments, however, these graphs are distributed across organizations and constrained by strict privacy policies, making centralized training infeasible. Federated learning provides a promising solution, yet its effectiveness is severely limited by the dual incompleteness of real-world semantic graphs: missing node attributes and incomplete relational structures. Such dual missingness, often heterogeneous and unobserved across clients, causes substantial degradation in model performance. We present FedCL, a missingness-aware federated contrastive learning framework for dual-incomplete semantic graphs. FedCL introduces two key components: a topology estimation module, grounded in rate–distortion theory, that privately quantifies structural incompleteness across clients, and a federated reconstruction module that leverages these estimations to generate plausible relations without inferring sensitive attributes. To further improve robustness, FedCL integrates graph contrastive learning across reconstructed subgraphs, ensuring semantic consistency across heterogeneous and incomplete client graphs. Experiments on benchmark citation and Web datasets demonstrate that FedCL consistently outperforms state-of-the-art baselines in accuracy and robustness under heterogeneous missingness, while preserving strong privacy guarantees. These results highlight FedCL as a scalable and trustworthy approach for federated learning on incomplete semantic graphs, advancing privacy-preserving knowledge sharing on the Web.
Recently, hypergraph knowledge distillation has been proposed to alleviate the high computational cost of Hypergraph Neural Networks (HGNNs) when modeling high-order relationships in Web-related graph tasks. Its effectiveness primarily depends on the quality of knowledge transferred from the teacher and the representation capability of the student. However, existing methods remain limited on both sides. On the teacher side, most methods typically rely on a single HGNN teacher, which provides limited structural and semantic knowledge, thereby constraining the upper bound of the student's performance. The potential of exploiting multiple teachers in HGNNs remains largely underexplored. On the student side, existing methods ignore the student's capability to capture high-order semantic and structural information beyond simply imitating teacher outputs, leading to limited representation learning. To address these limitations, we propose MARCH, a framework for Multi-TeAcheR Contrastive Hypergraph Distillation, which advances semantic modeling and distillation for Web-scale structured data. Specifically, MARCH proposes a multi-teacher distillation strategy that adaptively transfers complementary knowledge from multiple teachers at both node and hyperedge levels, empowering the student model to learn richer and more discriminative representations and even outperform its teachers. Extensive experiments on six benchmark datasets demonstrate the superior performance of MARCH.