Recent advances in knowledge representation learning (KRL) highlight the urgent necessity to unify symbolic knowledge graphs (KGs) with language models (LMs) for richer semantic understanding. However, existing approaches typically prioritize either graph structure or textual semantics, which leaves a gap, i.e., a unified framework that simultaneously captures global KG connectivity, nuanced linguistic context, and discriminative reasoning semantics. To bridge this gap, we introduce KG-BiLM, a bidirectional LM framework that fuses structural cues from KGs with the semantic expressiveness of generative transformers. KG-BiLM incorporates three key components: (i) Bidirectional Knowledge Attention, which eliminates the causal mask to enable full interaction among all tokens and entities; (ii) Knowledge-Masked Prediction, which encourages the model to leverage both local semantic contexts and global graph connectivity; and (iii) Contrastive Graph Semantic Aggregation, which preserves KG structure via contrastive alignment of sampled subgraph representations. Extensive experiments on standard benchmarks demonstrate that KG-BiLM outperforms the state-of-the-art baselines in link prediction, especially on large-scale graphs with complex multi-hop relations—validating its effectiveness in unifying structural information and textual semantics. The source code of KG-BiLM is available at https://github.com/zirui-chen/kg-bilm
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
Knowledge Hypergraphs (KHs) have recently emerged as a knowledge representation for retrieval-augmented generation (RAG), offering a paradigm to model multi-entity relations into a structured form. However, existing KH-based RAG methods suffer from three major limitations: static retrieval planning, non-adaptive retrieval execution, and superficial use of KH structure and semantics, which constrain their ability to perform effective multi-hop question answering. To overcome these limitations, we propose PRoH, a dynamic Planning and Reasoning over Knowledge Hypergraphs framework. PRoH incorporates three core innovations: (i) a context-aware planning module that sketches the local KH neighborhood to guide structurally grounded reasoning plan generation; (ii) a structured question decomposition process that organizes subquestions as a dynamically evolving Directed Acyclic Graph (DAG) to enable adaptive, multi-trajectory exploration; and (iii) an Entity-Weighted Overlap (EWO)-guided reasoning path retrieval algorithm that prioritizes semantically coherent hyperedge traversals. Experiments across multiple domains demonstrate that PRoH achieves state-of-the-art performance, surpassing the prior SOTA model HyperGraphRAG by an average of 19.73% in F1 and 8.41% in Generation Evaluation (G-E) score, while maintaining strong robustness in long-range multi-hop reasoning tasks.
Large Language Models (LLMs) possess strong representation and reasoning capabilities, but their application to structure-based drug design (SBDD) is limited by insufficient understanding of protein structures and unpredictable molecular generation. To address these challenges, we propose Exploration-Augmented Latent Inference for LLMs (ELILLM), a framework that reinterprets the LLM generation process as an encoding, latent space exploration, and decoding workflow. ELILLM explicitly explores portions of the design problem beyond the model's current knowledge while using a decoding module to handle familiar regions, generating chemically valid and synthetically reasonable molecules. In our implementation, Bayesian optimization guides the systematic exploration of latent embeddings, and a position-aware surrogate model efficiently predicts binding affinity distributions to inform the search. Knowledge-guided decoding further reduces randomness and effectively imposes chemical validity constraints. We demonstrate ELILLM on the CrossDocked2020 benchmark, showing strong controlled exploration and high binding affinity scores compared with seven baseline methods. These results demonstrate that ELILLM can effectively enhance LLMs' capabilities for SBDD. Our code is available at https://github.com/hxnhxn/ELILLM.
Accurate representation of multimodal knowledge is crucial for event forecasting in real-world scenarios. However, existing studies have largely focused on static settings, overlooking the dynamic acquisition and fusion of multimodal knowledge. 1) At the knowledge acquisition level, how to learn time-sensitive information of different modalities, especially the dynamic structural modality. Existing dynamic learning methods are often limited to shallow structures across heterogeneous spaces or simple unispaces, making it difficult to capture deep relation-aware geometric features. 2) At the knowledge fusion level, how to learn evolving multimodal fusion features. Existing knowledge fusion methods based on static coattention struggle to capture the varying historical contributions of different modalities to future events. To this end, we propose DyMRL, a Dynamic Multispace Representation Learning approach to efficiently acquire and fuse multimodal temporal knowledge. 1) For the former issue, DyMRL integrates time-specific structural features from Euclidean, hyperbolic, and complex spaces into a relational message-passing framework to learn deep representations, reflecting human intelligences in associative thinking, high-order abstracting, and logical reasoning. Pretrained models endow DyMRL with time-sensitive visual and linguistic intelligences. 2) For the latter concern, DyMRL incorporates advanced dual fusion-evolution attention mechanisms that assign dynamic learning emphases equally to different modalities at different timestamps in a symmetric manner. To evaluate DyMRL's event forecasting performance through leveraging its learned multimodal temporal knowledge in history, we construct four multimodal temporal knowledge graph benchmarks. Extensive experiments demonstrate that DyMRL outperforms state-of-the-art dynamic unimodal and static multimodal baseline methods.
Large Language Models (LLMs) have achieved impressive reasoning abilities, but struggle with temporal understanding, especially when questions involve multiple entities, compound operators, and evolving event sequences. Temporal Knowledge Graphs (TKGs), which capture vast amounts of temporal facts in a structured format, offer a reliable source for temporal reasoning. However, existing TKG-based LLM reasoning methods still struggle with four major challenges: maintaining temporal faithfulness in multi-hop reasoning, achieving multi-entity temporal synchronization, adapting retrieval to diverse temporal operators, and reusing prior reasoning experience for stability and efficiency. To address these issues, we propose MemoTime, a memory-augmented temporal knowledge graph framework that enhances LLM reasoning through structured grounding, recursive reasoning, and continual experience learning. MemoTime decomposes complex temporal questions into a hierarchical Tree of Time, enabling operator-aware reasoning that enforces monotonic timestamps and co-constrains multiple entities under unified temporal bounds. A dynamic evidence retrieval layer adaptively selects operator-specific retrieval strategies, while a self-evolving experience memory stores verified reasoning traces, toolkit decisions, and sub-question embeddings for cross-type reuse. Comprehensive experiments on multiple temporal QA benchmarks show that MemoTime achieves overall state-of-the-art results, outperforming the strong baseline by up to 24.0%. Furthermore, MemoTime enables smaller models (e.g., Qwen3-4B) to achieve reasoning performance comparable to that of GPT-4-Turbo.
Scholarly web is a vast network of knowledge connected by citations. However, this system is increasingly compromised by miscitation, where references do not support or even contradict the claims they are cited for. Current miscitation detection methods, which primarily rely on semantic similarity or network anomalies, struggle to capture the nuanced relationship between a citation's context and its place in the wider network. While large language models (LLMs) offer powerful capabilities in semantic reasoning for this task, their deployment is hindered by hallucination risks and high computational costs. In this work, we introduce LLM-Augmented Graph Learning-based Miscitation Detector (LAGMiD), a novel framework that leverages LLMs for deep semantic reasoning over citation graphs and distills this knowledge into graph neural networks (GNNs) for efficient and scalable miscitation detection. Specifically, LAGMiD introduces an evidence-chain reasoning mechanism, which uses chain-of-thought prompting, to perform multi-hop citation tracing and assess semantic fidelity. To reduce LLM inference costs, we design a knowledge distillation method aligning GNN embeddings with intermediate LLM reasoning states. A collaborative learning strategy further routes complex cases to the LLM while optimizing the GNN for structure-based generalization. Experiments on three real-world benchmarks show that LAGMiD achieves state-of-the-art miscitation detection with significantly reduced inference cost.
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.
Despite rapid development, large language models (LLMs) still encounter challenges in multi-turn decision-making tasks (i.e., agent tasks) like web shopping and browser navigation, which require making a sequence of intelligent decisions based on environmental feedback. Previous work for LLM agents typically relies on elaborate prompt engineering or fine-tuning with expert trajectories to improve performance. In this work, we take a different perspective: we explore constructing process reward models (PRMs) to evaluate each decision and guide the agent's decision-making process. Unlike LLM reasoning, where each step is scored based on correctness, actions in agent tasks do not have a clear-cut correctness. Instead, they should be evaluated based on their proximity to the goal and the progress they have made. Building on this insight, we propose a re-defined PRM for agent tasks, named AgentPRM, to capture both the interdependence between sequential decisions and their contribution to the final goal. This enables better progress tracking and exploration-exploitation balance. To scalably obtain labeled data for training AgentPRM, we employ a Temporal Difference-based (TD-based) estimation method combined with Generalized Advantage Estimation (GAE), which proves more sample-efficient than prior methods. Extensive experiments across different agentic tasks show that AgentPRM is over 8× more compute-efficient than baselines, and it demonstrates robust improvement when scaling up test-time compute. Moreover, we perform detailed analyses to show how our method works and offer more insights, e.g., applying AgentPRM to the reinforcement learning of LLM agents.
The integration of Domain Knowledge Graphs (DKG) into Retrieval-Augmented Generation has emerged as a promising approach for constructing Specialized Large Language Models (spLLMs).On account of the scarcity of high-quality DKGs, existing approaches employ an evolutionary framework, wherein the DKG is continuously evolved alongside its utilization for enhancing the LLM. Yet, these methods face two key limitations: 1) heavy reliance on costly expert knowledge, and 2) neglect of the connection between the LLM's inherent knowledge and external expert knowledge. To address these issues, this paper introduces Epistemic Cognition-enhanced Specialized LLMs (EC-spLLM), a novel evolutionary framework that instills epistemic cognition into the LLM to systematically exploit both its internal knowledge and expert knowledge. At the core of EC-spLLM lies the Hallucination Detection-based Epistemic Cognition (HDEC) mechanism, which assesses the reliability of LLM-generated responses using the LLM's self cognition and hallucination detection. This assessment ability enables EC-spLLM to selectively adopt either the expert-provided golden answer or a reliable LLM-generated answer during DKG evolution, thereby reducing dependence on experts and bridging internal and external knowledge sources to enhance performance. We conducted extensive experiments on five datasets spanning five domain, e.g., emotional sociology, biology, ect. Results show that EC-spLLM reduces the usage of golden answers by an average of 67% while retaining 97.2% of the accuracy achieved by the SOTA method, and outperforms all other baselines.
The evaluation of Kleene Closure Property Path Queries (KPPQs) over RDF graphs presents a fundamental trade-off. Under the intuitive simple path semantics, which forbids node repetition, the problem is NP-complete. This forces practical systems to adopt existential path semantics, which admits tractable evaluation but produces result sets bloated with redundant and often semantically meaningless paths due to unnecessary detours through cycles. This paper introduces a novel indexing framework that effectively navigates this trade-off. Our key insight is that while enumerating all simple paths is intractable, a carefully selected, compact set of representative simple paths is sufficient to correctly and completely answer connectivity queries in practice. We propose the Node Clustered Index (NCI), which pre-materializes such paths—including the longest acyclic chains and fundamental cycles —transforming online query processing from a costly graph traversal into an efficient index lookup and expansion operation. Crucially, our method guarantees that every result conforms to simple path semantics. Extensive experiments demonstrate that our approach achieves near-perfect recall, outperforming state-of-the-art systems by up to an order of magnitude. This work establishes that for NP-complete path problems, seeking polynomial-time, empirically perfect approximations is a viable and powerful paradigm, delivering both performance and quality without compromise.
Cross-domain sequential recommendation (CDSR) plays a critical role in decentralized Web applications by leveraging user behavior sequences across multiple platforms to alleviate data sparsity and capture dynamic preferences. However, existing federated CDSR frameworks face two fundamental challenges: (i) heterogeneous sequential interactions that encode domain-exclusive semantics and cannot be directly shared under privacy constraints, and (ii) strong trust assumptions that both servers and clients behave honestly, leaving federated training vulnerable to misreporting, malicious updates, and negative transfer. In this paper, we propose VeriFRL, a verifiable federated representation learning framework for cross-domain sequential recommendation. VeriFRL adopts a dual-module design that integrates representation learning with verifiable training: an attention-based variational encoder disentangles domain-shared and domain-exclusive representations to support transferable and privacy-preserving knowledge sharing, while a contribution evaluation module quantifies client-level and feature-level influences to enable verifiability, interpretability, and negative transfer detection. Extensive experiments on real-world multi-domain datasets demonstrate that VeriFRL achieves competitive or superior recommendation performance over state-of-the-art federated CDSR methods, while providing fine-grained insights into cross-domain knowledge transfer dynamics.
Knowledge editing enables multimodal large language models (MLLMs) to efficiently update outdated or incorrect information. However, existing benchmarks primarily emphasize cognitive-level modifications while lacking a focus on deeper meta-cognitive processes. To bridge this gap, we introduce CogEdit, a novel benchmark designed to evaluate MLLMs' meta-cognitive knowledge editing abilities across three levels: (1) Counterfactual-Driven Editing, assessing self-awareness of knowledge correctness changes; (2) Boundary Constraint Editing, ensuring appropriate generalization without unintended interference; and (3) Noise-Robust Editing, promoting reflective evaluation of uncertain information. To advance meta-cognitive editing, we propose MIND (Meta-cognitive INtegrated Dynamic Knowledge Editing), a framework that constructs a meta-knowledge memory for self-awareness, employs game-theoretic interactions to monitor knowledge activation, and incorporates label refinement for noise-robust updates. Extensive experiments show that MIND significantly outperforms existing cognitive editing approaches, achieving strong performance on both traditional and meta-cognitive knowledge editing benchmarks.
Large-scale multimodal contrastive learning has recently achieved impressive success in learning rich and transferable representations, yet it remains fundamentally limited by the uniform treatment of feature dimensions and the neglect of the intrinsic spectral structure of the learned features. Empirical evidence indicates that high-dimensional embeddings tend to collapse into narrow cones, concentrating task-relevant semantics in a small subspace, while the majority of dimensions remain occupied by noise and spurious correlations. Such spectral imbalance and entanglement undermine model generalization. We propose Spectral Disentanglement and Enhancement (SDE), a novel framework that bridges the gap between the geometry of the embedded spaces and their spectral properties. Our approach leverages singular value decomposition to adaptively partition feature dimensions into strong signals that capture task-critical semantics, weak signals that reflect ancillary correlations, and noise representing irrelevant perturbations. A curriculum-based spectral enhancement strategy is then applied, selectively amplifying informative components with theoretical guarantees on training stability. Building upon the enhanced features, we further introduce a dual-domain contrastive loss that jointly optimizes alignment in both the feature and spectral spaces, effectively integrating spectral regularization into the training process and encouraging richer, more robust representations. Extensive experiments on large-scale multimodal benchmarks demonstrate that SDE consistently improves representation robustness and generalization, outperforming state-of-the-art methods. SDE integrates seamlessly with existing contrastive pipelines, offering an effective solution for multimodal representation learning.
Metaphors are a fundamental cognitive tool for articulating abstract and subjective experiences and implicit semantics, making them potent indicators of psychological state, particularly in individuals with depression. The proliferation of social media has created a vast repository of such metaphorical expressions, offering an unprecedented opportunity to understand mental health struggles. These metaphors can provide crucial insights for clinical assessment and therapeutic intervention. However, their potential remains largely untapped in automated depression detection, primarily due to the lack of large-scale, annotated datasets. To bridge this gap, we introduce the Depression-Related Metaphor Dataset (DRMD), a novel resource of social media posts related to depression, incorporating depression levels (severe, moderate, minimum, and null), the presence or absence of metaphors, and their conceptual source domain mappings. We leverage this dataset to fine-tune Large Language Models (LLMs), integrating metaphorical features to enhance detection capabilities. Our results demonstrate that models incorporating metaphorical information achieve superior accuracy in depression detection and, importantly, generate high-quality explanations for their decisions by referencing specific metaphorical expressions. This work underscores the critical role of metaphorical analysis in computational mental health and provides a foundation for future research in explainable AI for depression detection. The dataset is publicly available.
Knowledge Tracing (KT) is pivotal in intelligent tutoring systems, as it models the dynamic evolution of student knowledge from their learning interactions. However, the cross-disciplinary generalization of existing KT models is subjected to a dual constraint: the heterogeneity in students' cognitive abilities and the divergent disciplinary-specific knowledge structures.To address this challenge, we propose DTransKT, a dual transferable knowledge tracing framework tailored for cross-disciplinary adaptability. DTransKT enhances existing knowledge tracing models by dynamically aligning student representations and integrating external knowledge semantics.Specifically, the framework incorporates a Cross-disciplinary Graph-matching (CG) module, which captures meta-skill representations based on students' learning trajectories. Through cross-disciplinary node matching, the CG module aligns student-specific features, thereby improving tracing accuracy. Additionally, the Cross-disciplinary Attention-assisting (CA) module leverages pre-trained language models to extract meta-semantic from textual content, enhancing transferability.Extensive experimental evaluations demonstrate that DTransKT consistently enhances the performance of seven prominent KT models under direct transfer settings, achieving average improvements of 14.2% in accuracy (ACC) and 4.5% in area under the curve (AUC) across diverse datasets. These findings affirm the efficacy of our approach in enabling cross-disciplinary transfer for knowledge tracing. Code and pre-trained models are available at: https://github.com/Dual-KT/DTransKT.
Cognitive diagnostics in the Web-based Intelligent Education System (WIES) aims to assess students' mastery of knowledge concepts from heterogeneous, noisy interactions. Recent work has tried to utilize Large Language Models (LLMs) for cognitive diagnosis, yet LLMs struggle with structured data and are prone to noise-induced misjudgments. Specially, WIES's open environment continuously attracts new students and produces vast amounts of response logs, exacerbating the data imbalance and noise issues inherent in traditional educational systems. To address these challenges, we propose DLLM, a Diffusion-based LLM framework for noise-robust cognitive diagnosis. DLLM first constructs independent subgraphs based on response correctness, then applies relation augmentation alignment module to mitigate data imbalance. The two subgraph representations are then fused and aligned with LLM-derived, semantically augmented representations. Importantly, before each alignment step, DLLM employs a two-stage denoising diffusion module to eliminate intrinsic noise while assisting structural representation alignment. Specifically, unconditional denoising diffusion first removes erroneous information, followed by conditional denoising diffusion based on graph signal to eliminate misleading information. Finally, the noise-robust representation that integrates semantic knowledge and structural information is fed into existing cognitive diagnosis models for prediction. Experimental results on three publicly available web-based educational platform datasets demonstrate that our DLLM achieves optimal predictive performance across varying noise levels, which demonstrates that DLLM achieves noise robustness while effectively leveraging semantic knowledge from LLM.
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.
The Web is a rich source of structured data in the form of tables, from product catalogs and knowledge bases to scientific datasets. However, the heterogeneity of the structure and semantics of these tables makes it challenging to build a unified method that can effectively leverage the information they contain. Meanwhile, Large language models (LLMs) are becoming an increasingly integral component of web infrastructure for tasks like semantic search. This raises a crucial question: can we leverage these already-deployed LLMs to classify structured data in web-native tables (e.g., product catalogs, knowledge base exports, scientific data portals), avoiding the need for specialized models or extensive retraining? This work investigates a lightweight paradigm, Table Representation with Language Model (TaRL), for few-shot tabular classification that directly utilizes semantic embeddings of individual table rows. We first show that naive application of these embeddings underperforms compared to specialized tabular models. We then demonstrate that their potentials can be unlocked with two key techniques: removing the common component from all embeddings and calibrating the softmax temperature. We show that a simple meta-learner, trained on handcrafted features, can learn to predict an appropriate temperature. This approach achieves performance comparable to state-of-the-art models in low-data regimes (k ? 32) of semantically-rich tables. Our findings demonstrate the viability of reusing existing LLM infrastructure for efficient semantics-driven pathway to reuse existing LLM infrastructure for Web table understanding.