Retrieval-Augmented Generation (RAG) has emerged as a paradigm for enhancing large language models (LLMs) with external knowledge, yet existing graph-based methods face a fundamental limitation: entity-centric and chunk-centric approaches operate on representations anchored to original text without true knowledge fusion. While entity-centric methods connect logically related content and chunk-centric methods preserve context, both retrieve information separately through similarity search, missing emergent understanding from their synthesis. In this paper, we propose HyGRAG, a hierarchical graph RAG framework that transcends source documents by addressing three core challenges: constructing summaries that genuinely integrate contextual and relational information, leveraging these synthesized representations to access emergent knowledge during retrieval, and efficiently updating hierarchical structures for dynamic corpora. Specifically, we design hierarchical index structures over hybrid graphs with both chunk and entity nodes, then iteratively cluster them and generate LLM-based summaries. Then, we design context and relation-aware retrieval that searches across all abstraction levels while expanding through community membership. Moreover, we enable dynamic knowledge update through attachment-based algorithms with only local re-summarization. Experimental results show that HyGRAG improves the average accuracy of multi-hop reasoning tasks by 9.7%, while maintaining reasonable efficiency.
论文检索
输入标题、作者或关键词，从 2,942 篇学术成果中精准定位
SketchMind: Understanding Abstract Sketches with MLLMs for Fine-Grained Sketch-Based Image Retrieval
Fine-Grained Sketch-Based Image Retrieval (FG-SBIR) aims to retrieve images that accurately correspond to abstract hand-drawn sketches, requiring the model to understand sparse and abstract visual cues. Existing methods tend to rely on convolutional networks or metric learning to align sketch and image features, often overlooking the inherent abstraction and semantic ambiguity present in sketches. This limitation results in an insufficient understanding of fine-grained visual details. To address this challenge, we propose SketchMind, a novel method that leverages Multi-modal Large Language Models (MLLMs) to enhance abstract sketch understanding in FG-SBIR. Specifically, we use MLLMs to generate auxiliary textual descriptions based on the given sketches via a Visual Question Answering (VQA) strategy. To effectively incorporate these descriptions, we construct a graph structure with the sketch as the central node and the generated texts as peripheral nodes. A graph attention scheme is employed to perform uncertainty-aware feature fusion, enabling the model to suppress noisy or irrelevant textual information. Furthermore, to enhance both inter- and intra-modal fine-grained alignment, we design a Multi-scale Cross-modal Jigsaw Matching module in combination with a self-supervised learning strategy, which captures local and global visual correspondences across modalities more effectively. Extensive experiments on three benchmark FG-SBIR datasets demonstrate that SketchMind achieves superior performance over existing state-of-the-art methods, proving its effectiveness. Code is available at https://github.com/li1changxing/MLLM_FG_SBIR/.
Scientific document retrieval is a critical task for enabling knowledge discovery and supporting research across diverse domains. However, existing dense retrieval methods often struggle to capture fine-grained scientific concepts in texts due to their reliance on holistic embeddings and limited domain understanding. Recent approaches leverage large language models (LLMs) to extract fine-grained semantic entities and enhance semantic matching, but they typically treat entities as independent fragments, overlooking the multi-faceted nature of scientific concepts. To address this limitation, we propose Pairwise Semantic Matching (PairSem), a framework that represents relevant semantics as entity–aspect pairs, capturing complex, multi-faceted scientific concepts. PairSem is unsupervised, base retriever-agnostic, and plug-and-play, enabling precise and context-aware matching without requiring query-document labels or entity annotations. Extensive experiments on multiple datasets and retrievers demonstrate that PairSem significantly improves retrieval performance, highlighting the importance of modeling multi-aspect semantics in scientific information retrieval.
Despite extensive research on a wide range of question answering (QA) systems, most existing work focuses on answer containment—i.e., assuming that answers can be directly extracted and/or generated from documents in the corpus. However, some questions require inference, i.e., deriving answers that are not explicitly stated but can be inferred from the available information. We introduce Inferential QA–a new task that challenges models to infer answers from answer-supporting passages which provide only clues. To study this problem, we construct Quit (QUestions requiring Inference from Texts) dataset, comprising 7,401 questions and 2.4M passages built from high-convergence human- and machine-authored hints, labeled across three relevance levels using LLM-based answerability and human verification. Through comprehensive evaluation of retrievers, rerankers, and LLM-based readers, we show that methods effective on traditional QA tasks struggle in inferential QA: retrievers underperform, rerankers offer limited gains, and fine-tuning provides inconsistent improvements. Even reasoning-oriented LLMs fail to outperform smaller general-purpose models. These findings reveal that current QA pipelines are not yet ready for inference-based reasoning. Inferential QA thus establishes a new class of QA tasks that move towards understanding and reasoning from indirect textual evidence.
Cross-modal retrieval is a fundamental task in multimedia understanding, aimed at querying samples with similar semantics in one modality (e.g., text) using another modality (e.g., image). Existing methods merely focus on point-to-point comparisons between individual samples, while overlooking the widely present many-to-many structural relationships in real-world scenarios. However, the many-to-many relationships formed by multiple samples sharing similar semantics are crucial for effectively achieving semantic alignment and accurately constructing shared semantic representations. To address this, we propose a novel hypergraph-based cross-modal retrieval approach, which explicitly establishes many-to-many associations between multiple samples using a label-driven hypergraph construction mechanism, combined with differentiated hyperedge weighting. Additionally, to avoid the limitation of information interaction direction imposed by traditional unidirectional cross-attention mechanisms, we design a bidirectional cross-attention structure, with image and text as separate query sources, to achieve symmetric semantic enhancement between modalities. The resulting joint image-text representations are then mapped as hypergraph vertices, further enhancing the model's ability to align cross-modal semantics. Since constructing a global hypergraph on a large-scale sample set would incur high computational cost, we introduce global label co-occurrence frequency to supervise the batch-level hypergraph construction, enhancing the local graph's ability to capture global semantics. Experimental results show that our model outperforms existing state-of-the-art methods on three benchmark cross-modal retrieval datasets.
The effectiveness upper bound of retrieval-augmented generation (RAG) is fundamentally constrained by the semantic integrity and information granularity of text chunks in its knowledge base. Moreover, domain documents are characterized by dense terminology and strong contextual dependencies, which exacerbate the semantic fragmentation of text chunks, thereby making it difficult to efficiently utilize their key information. To address these challenges, this paper proposes QChunker, which restructures the RAG paradigm from retrieval-augmentation to understanding-retrieval-augmentation. Firstly, QChunker models the text chunking as a composite task of text segmentation and knowledge completion to ensure the logical coherence and integrity of text chunks. Drawing inspiration from Hal Gregersen's ''Questions Are the Answer'' theory, we design a multi-agent debate framework comprising four specialized components: a question outline generator, text segmenter, integrity reviewer, and knowledge completer. This framework operates on the principle that questions serve as catalysts for profound insights. Through this pipeline, we successfully construct a high-quality dataset of 45K entries and transfer this capability to small language models. Additionally, to handle long evaluation chains and low efficiency in existing chunking evaluation methods, which overly rely on downstream QA tasks, we introduce a novel direct evaluation metric, ChunkScore. Both theoretical and experimental validations demonstrate that ChunkScore can directly and efficiently discriminate the quality of text chunks. Furthermore, during the text segmentation phase, we utilize document outlines for multi-path sampling to generate multiple candidate chunks and select the optimal solution employing ChunkScore. Extensive experimental results across four heterogeneous domains exhibit that QChunker effectively resolves aforementioned issues by providing RAG with more logically coherent and information-rich text chunks. Notably, this study also establishes a small-domain QA dataset concerning hazardous chemical safety, which fully reveals the significant value of RAG in specialized domains and the generalization capability of the QChunker framework.
Composed Image Retrieval (CIR) aims to retrieve a target image from a query composed of a reference image and modification text. Recent training-free zero-shot methods often employ Multimodal Large Language Models (MLLMs) with Chain-of-Thought (CoT) to compose a target image description for retrieval. However, due to the fuzzy matching nature of ZS-CIR, the generated description is prone to semantic bias relative to the target image. We propose SDR-CIR, a training-free Semantic Debias Ranking method based on CoT reasoning. First, Selective CoT guides the MLLM to extract visual content relevant to the modification text during image understanding, thereby reducing visual noise at the source. We then introduce a Semantic Debias Ranking with two steps, Anchor and Debias, to mitigate semantic bias. In the Anchor step, we fuse reference image features with target description features to reinforce useful semantics and supplement omitted cues. In the Debias step, we explicitly model the visual semantic contribution of the reference image to the description and incorporate it into the similarity score as a penalty term. By supplementing omitted cues while suppressing redundancy, SDR-CIR mitigates semantic bias and improves retrieval performance. Experiments on three standard CIR benchmarks show that SDR-CIR achieves state-of-the-art results among one-stage methods while maintaining high efficiency. The code is publicly available at https://github.com/suny105/SDR-CIR.
Existing multimodal document question-answering (QA) systems predominantly rely on flat semantic retrieval, representing documents as a set of disconnected text chunks and largely neglecting their intrinsic hierarchical and relational structures. Such flattening disrupts logical and spatial dependencies—such as section organization, figure-text correspondence, and cross-reference relations—that humans naturally exploit for comprehension. To address this limitation, we introduce a document-level structural Document MAP (DMAP), which explicitly encodes both hierarchical organization and inter-element relationships within multimodal documents. Specifically, we design a Structured-Semantic Understanding Agent to construct DMAP by organizing textual content together with figures, tables, charts, etc into a human-aligned hierarchical schema that captures both semantic and layout dependencies. Building upon this representation, a Reflective Reasoning Agent performs structure-aware and evidence-driven reasoning, dynamically assessing the sufficiency of retrieved context and iteratively refining answers through targeted interactions with DMAP. Extensive experiments on MMDocQA benchmarks demonstrate that DMAP yields document-specific structural representations aligned with human interpretive patterns, substantially enhancing retrieval precision, reasoning consistency, and multimodal comprehension over conventional RAG-based approaches. Code is available at https://github.com/Forlorin/DMAP
Long videos contain a vast amount of information, making video-text retrieval an essential and challenging task in multimodal learning and web-scale search. On today's Web, where users increasingly expect to locate not only relevant pages but also specific long videos or fine-grained clips, existing benchmarks fall short due to limited video duration, low-quality captions, and coarse annotation granularity. To address these limitations, we introduce LoVR, a benchmark specifically designed for long video-text retrieval. LoVR contains 467 long videos and over 40,804 fine-grained clips with high-quality captions. To overcome the issue of poor machine-generated annotations, we propose an efficient caption generation framework that integrates VLM automatic generation, caption quality scoring, and dynamic refinement. This pipeline improves annotation accuracy while maintaining scalability. Furthermore, we introduce a semantic fusion method to generate coherent full-video captions without losing important contextual information. Our benchmark introduces longer videos, more detailed captions, and a larger-scale dataset, presenting new challenges for video understanding and retrieval. Extensive experiments on various advanced models demonstrate that LoVR is a challenging benchmark, revealing the limitations of current approaches and providing valuable insights for future research. We release the code link at https://github.com/TechNomad-ds/LoVR-benchmark/.
Recent studies show that claims incorporating both text and images spread more effectively than those with text alone, presenting significant challenges for multimodal fact-checking. The rapid development of Multi-modal Large Language Models (MLLMs) has greatly advanced research in this field, enabling stronger performance. However, existing MLLM-based fact-checking methods fail to fully exploit visual evidence, and their reliance on rigid fine-tuning templates limits context-aware explanations and leads to weak deep reasoning. To address these limitations, we propose FACTCOMPASS, a novel framework that combines reasoning-aware fine-tuning with large-scale rule-based reinforcement learning and incorporates a semantic- and knowledge-enhanced retrieval module to strengthen deep reasoning and improve evidence utilization. This framework enhances evidence retrieval by obtaining semantically relevant evidence images, enriching the contextual understanding of claim-related images, and refining textual evidence at the knowledge level. To further enhance reasoning, we introduce a self-refining reinforcement fine-tuning strategy: (1) distilling GPT-4o's reasoning from partially fact-checking data for cold-start Chain-of-Thought learning; (2) activating reasoning across broader datasets using prior knowledge and rejection sampling; (3) applying Group Relative Policy Optimization to explore diverse reasoning paths and optimize factual consistency. Extensive experiments have demonstrated the effectiveness of the proposed framework.
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.
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.
Agricultural landscape segmentation in the Global South is challenging as it is characterized by fragmented plots, high intra-class variance, and a scarcity of labeled training data. Recent advances in segmentation have been made by Multimodal Large Language Models (MLLMs). However, current approaches encounter critical context length bottlenecks and a domain alignment gap in understanding satellite features. We address these limitations through MAgSeg, a novel, decoder-free MLLM segmentation approach. MAgSeg is an architecturally efficient approach that enables standard MLLMs to perform segmentation of complex smallholder agricultural landscapes from high-resolution satellite imagery, without requiring auxiliary vision decoders. We introduce a novel instruction tuning data format designed to enable scalable fine-tuning and post-training on high resolution satellite imagery, which enables MAgSeg to learn from the global context of the image while generating text tokens for only a patch within the image. Extensive evaluations on datasets spanning three countries in the Global South demonstrate that MAgSeg significantly outperforms state-of-the-art MLLM baselines, offering a scalable solution to map smallholder agricultural environments.
Multimodal meme understanding is increasingly used to analyze socially sensitive content, yet existing models often exhibit biased behavior when interpreting economic dependence and social roles under ambiguity. Many memes express economic relationships through sparse text or symbolic visual cues, providing insufficient evidence for gendered attribution. In such underspecified settings, models tend to rely on pretraining correlations, leading to hallucinated and stereotypical economic role assignments. In this work, we study gendered economic dependence in image-text memes through the lens of contextual sufficiency and identify epistemic overcommitment—inferring roles without adequate evidence—as a primary source of bias. We propose CGER-Net, a context-grounded multimodal framework that estimates whether the input provides sufficient evidence for gendered economic reasoning and applies evidence-gated inference to enable confident attribution when cues are explicit while favoring principled abstention otherwise. We evaluate CGER-Net on EconMeme-GE, a curated dataset of image-text memes annotated as Men, Women, Neutral, or Ambiguous. Across strong contemporary multimodal baselines, CGER-Net reduces Gender Overcommitment Rate by up to 44% on ambiguous instances while maintaining comparable accuracy on unambiguous cases. Human evaluation further shows that 79% of generated rationales are judged as epistemically aligned with the available evidence. These results highlight the importance of modeling when not to infer for reliable and responsible multimodal analysis.
Robots that follow natural-language instructions often either plan at a high level using hand-designed interfaces or rely on large end-to-end models that are difficult to deploy for real-time control. We propose TeNet (Text-to-Network), a framework for instantiating compact, task-specific robot policies directly from natural language descriptions. TeNet conditions a hypernetwork on text embeddings produced by a pretrained large language model (LLM) to generate a fully executable policy, which then operates solely on low-dimensional state inputs at high control frequencies. By using the language only once at the policy instantiation time, TeNet inherits the general knowledge and paraphrasing robustness of pretrained LLMs while remaining lightweight and efficient at execution time. To improve generalization, we optionally ground language in behavior during training by aligning text embeddings with demonstrated actions, while requiring no demonstrations at inference time. Experiments on MuJoCo and Meta-World benchmarks show that TeNet produces policies that are orders of magnitude smaller than sequence-based baselines, while achieving strong performance in both multi-task and meta-learning settings and supporting high-frequency control. These results show that text-conditioned hypernetworks offer a practical way to build compact, language-driven controllers for ressource-constrained robot control tasks with real-time requirements.
Vision-language-action models (VLAs) often achieve high performance on demonstrated tasks but struggle significantly when required to extrapolate, recombining skills used in different tasks in novel ways. For instance, VLAs might successfully put the cream cheese in the bowl and put the bowl on top of the cabinet, yet still fail to put the cream cheese on top of the cabinet. This motivates us to investigate whether VLAs merely overfit to demonstrated tasks or still hold the potential to extrapolate. Our study uses text latent as the ingredient; it is a task-specific vector derived from the models’ hidden states. It thus encodes semantics necessary for completing a task and can be used to reconstruct the associated task behavior by writing it to the model’s residual stream. Furthermore, we find that skills used in distinct tasks can be combined to produce novel behaviors by blending their respective text latent. Applying this to π0, we increase its success rate from 9% to 83% on the proposed libero-ood benchmark, which features 20 tasks extrapolated from standard LIBERO tasks. This reveals that the skill representations encoded in text-latent are individual yet composable, while π0 fails to autonomously combine these representations for extrapolation. This also validates the design of libero-ood; it comprises tasks that the model fails, yet should be able to complete. We then tested other VLAs on libero-ood, and none of them achieved a success rate higher than 21%. Further analysis reveals VLAs share a common pattern to exhibit spatial overfitting, associating object names with where the object is spatially located in the demonstrated scene rather than achieving true object and goal understanding.
Large Language Models (LLMs) are extensively used at biomedical text processing but often fail to capture the complex, functional relationships encoded in expert knowledge graphs like the Human Phenotype Ontology (HPO). This "semantic gap'" limits their utility in precision medicine tasks such as rare disease diagnosis, where distinguishing overlapping clinical presentations requires understanding underlying pathophysiological connections rather than just surface-level textual similarity. In this work, we propose a Neuro-Symbolic Alignment Framework that bridges this separation by integrating literature-mined specialized phenotypical descriptions with the ontological structure used as reference. Specifically, we augment phenotype representations with automatically selected text fragments from massive corpus of descriptions mined from scientific literature (PubMed), overcoming the typical data scarcity of standard ontology definitions. We define a new embedding adaptation procedure whose fine-tuning approach is guided by a novel "Disease-Overlap" similarity measure, which prioritizes clinical co-occurrence of phenotypes over taxonomic distance, and optimizes the embedding space using AnglE Loss to mitigate gradient saturation. Extensive evaluations show that our approach significantly outperforms state-of-the-art baselines, including SapBERT, on both intrinsic semantic correlation and practical downstream tasks, including synthetic patient disease ranking and solving real cases stored in Phenopacket, where our model achieves x4 top-1 accuracy than the previous best model.
Abstractive summarization plays a crucial role in enabling efficient understanding of scientific literature, yet it inherently demands both linguistic fluency and factual faithfulness. Existing approaches often fail to reconcile these two requirements. Extractive methods rely on rigid sentence splicing that disrupts macro-level logical coherence, while large language model (LLM)-based generative approaches, despite mastering linguistic fluency, exhibit limited factual consistency. In this work, we propose ScholarSum, a hierarchical reflective graph-based framework that emulates a student–teacher writing process for fluent and faithful scientific summarization. ScholarSum first organizes the document into a hierarchical knowledge graph by segmenting it into semantically coherent units, whose multi-layered community structure captures global logic and macro-level themes. Guided by this global structure, the student generates an initial draft, which is subsequently refined through fine-grained evidence retrieval. To ensure factual consistency, a teacher-like reviewer then iteratively examines the draft, identifies unsupported content, and prompts targeted re-retrieval and rewriting until the summary meets rigorous quality standards. Extensive experiments demonstrate that ScholarSum significantly outperforms previous baselines in terms of both completeness and faithfulness. Our code is available at https://github.com/Xiaoyu-Tao/ScholarSum.
While Vision-Language Models (VLMs) have shown promise in textual understanding, they face significant challenges when handling long context and complex reasoning tasks. In this paper, we dissect the internal mechanisms governing long-context processing in VLMs to understand their performance bottlenecks. Through the lens of attention analysis, we identify specific Visual Evidence Retrieval (VER) Heads —a sparse, dynamic set of attention heads critical for locating visual cues during reasoning, distinct from static OCR heads. We demonstrate that these heads are causal to model performance; masking them leads to significant degradation. Leveraging this discovery, we propose VERA (Visual Evidence Retrieval Augmentation), a training-free framework that detects model uncertainty (i.e., entropy) to trigger the explicit verbalization of visual evidence attended by VER heads. Comprehensive experiments demonstrate that VERA significantly improves long-context understanding of open-source VLMs: it yields an average relative improvement of 21.3% on Qwen3-VL-8B-Instruct and 20.1% on GLM-4.1V-Thinking across five benchmarks.
Structured claim decomposition is often proposed as a solution for verifying complex, multi-faceted claims, yet empirical results have been inconsistent. We argue that these inconsistencies stem from two overlooked bottlenecks: evidence alignment and sub-claim error profiles. To better understand these factors, we introduce a new dataset of real-world complex claims, featuring temporally bounded evidence and human-annotated sub-claim evidence spans. We evaluate decomposition under two evidence alignment setups: Sub-claim Aligned Evidence (SAE) and Repeated Claim-level Evidence (SRE). Our results reveal that decomposition brings significant performance improvement only when evidence is granular and strictly aligned. By contrast, standard setups that rely on repeated claim-level evidence (SRE) fail to improve and often degrade performance as shown across different datasets and domains (PHEMEPlus, MMM-Fact, COVID-Fact). Furthermore, we demonstrate that in the presence of noisy sub-claim labels, the nature of the error ends up determining downstream robustness. We find that conservative "abstention" significantly reduces error propagation compared to aggressive but incorrect predictions. These findings suggest that future claim decomposition frameworks must prioritize precise evidence synthesis and calibrate the label bias of sub-claim verification models.