literature synthesis

To address time consumption, reading fatigue, and analytical challenges faced by STEM researchers when processing peer review feedback, this paper proposes a contextualized interactive support paradigm and develops a context-aware, visualization-guided system for reviewing feedback comprehension. Methodologically, it integrates user interviews, storyboard-based narrative design, and human-computer interaction principles—overcoming limitations of existing frameworks, including insufficient domain-specific theoretical guidance and the lack of domain-aware NLP tools. A controlled experiment (N=31) demonstrates significant improvements over conventional approaches in comprehension efficiency, integration accuracy, and cognitive load reduction. Field deployment (N=6) further validates its effectiveness and high usability in real-world academic workflows. The core contribution is the first application of contextualized interaction to scholarly review feedback processing, establishing a scalable, interpretable, and context-rich human-AI collaboration paradigm.

To address the problem of ambiguous reviewer comments in peer review—leading to difficulties in author comprehension, prolonged revision cycles, and limited manuscript improvement—this paper proposes a dual-graph collaborative modeling framework. It constructs a semantic mental graph (explicitly modeling the author’s cognitive pathway) and a hierarchical background graph (structurally embedding domain-specific knowledge), jointly optimized via graph neural network–driven retrieval for intent-level feedback interpretation. This work is the first to unify and co-reason over the author’s thought process and domain knowledge structure. Experiments demonstrate significant improvements: +23.6% in key-point identification accuracy and +31.4% in explanation consistency over prior state-of-the-art methods across multiple benchmark tasks. Empirical evaluation further shows an average 37% reduction in revision cycle time, enhancing both transparency and actionability of peer review feedback.

Existing research ideation tools emphasize breadth-oriented idea generation but lack support for iterative refinement, elaboration, and evaluation—hindering literature-grounded, deep-reading–driven conceptual evolution. Method: We propose the first literature-driven interactive research ideation system, integrating a composable “idea element” canvas model with a multi-dimensional (problem/solution/evaluation/contribution) co-evolution mechanism. Our approach innovatively incorporates LLM-powered literature-aware feedback generation, graph-structured idea modeling, and interactive multi-path variant exploration. Contribution/Results: Experiments demonstrate a 42% increase in user-generated idea output and significantly enhanced detail elaboration. Seven researchers successfully applied the system across the full ideation pipeline—from initial topic conception to paper outline revision—validating its efficacy in supporting deep, iterative, literature-informed research design.

To address the limitation of existing reviewer-paper matching methods in academic conferences—namely, their overreliance on a single factor leading to biased and incomplete assessments—this paper proposes a multi-factor joint modeling framework. Methodologically, it introduces the novel “factor-chain reasoning” paradigm, which decouples semantic, topical, and citation-based relevance signals into modular, composable, and interpretable components. These modules are orchestrated via instruction-tuned, context-aware language models to generate domain-agnostic scientific text embeddings, while a chain-structured architecture enables dynamic weight adjustment and progressive candidate filtering. Evaluated across four major domains—including machine learning and computer vision—the framework achieves significant improvements over state-of-the-art methods. On a newly constructed benchmark dataset, it attains a 12.7% absolute gain in mean Average Precision (mAP) and a Top-5 recall rate of 91.4%.

To address low literature review efficiency, high domain-knowledge barriers, and prominent hallucination risks in large language models (LLMs), this work proposes a fully automated LLM-based literature review generation method. We introduce a statistically validated, multi-layer quality control framework that reduces factual hallucination to below 0.5% (95% confidence interval), while ensuring citation completeness and enabling domain-agnostic prompt engineering. The system supports cross-disciplinary deployment without requiring domain expertise and is distributed as a one-click Windows desktop application. Empirical evaluation in propane dehydrogenation catalysis demonstrates its efficacy: it processes 343 papers in seconds, covering 35 thematic categories; extended analysis of 1,041 papers achieves expert-validated accuracy and citation integrity.

Existing approaches struggle to assess, at a fine-grained level, how citations in interdisciplinary research substantively integrate ideas from multiple fields. This work proposes a citation-purpose classification framework tailored for interdisciplinary scholarship, combining manual annotation, citation context analysis, and qualitative categorization to construct the first quantifiable system that measures both the depth and significance of citation engagement. Validated on publications at the intersection of natural language processing and computational social science, the framework not only uncovers actual patterns of interdisciplinary citation usage but also provides an actionable metric for evaluating citation quality.

This work addresses the challenge researchers often face in balancing novelty with effective grounding in existing literature when developing new ideas, as well as the lack of tools that support dynamic interaction between emerging concepts and relevant scholarly works. The paper introduces a novel “literature-driven idea pivoting” mechanism—a closed-loop framework that integrates idea drafting, dynamic literature retrieval, semantic clustering, and generative critical feedback to enable co-evolution of research ideas and the literature space. The system performs context-aware analysis of partial idea content and provides real-time improvement suggestions based on clusters of relevant papers. Experimental results demonstrate that this approach significantly enhances the quality of user-generated ideas and strengthens researchers’ ability to comprehend and leverage the scholarly context effectively.

Peer review is experiencing exponential growth in volume, yet its quality remains highly variable, necessitating systematic, interpretable, and scalable evaluation tools. This work proposes PeeriScope—the first modular platform that integrates structured features, rubric-based large language model assessments, and supervised learning predictions to enable multidimensional, explainable quantification of review quality. PeeriScope supports diverse use cases including self-assessment by reviewers, editorial screening, and large-scale audit studies. Designed for real-world deployment and research extensibility, the platform offers an open API and a web interface. The project is open-sourced and accompanied by an online demo, aiming to foster continuous innovation and practical adoption of robust peer review evaluation methodologies.

Scientific literature exhibits high heterogeneity, and manual meta-analyses are inefficient and error-prone. Method: This paper proposes an intelligent agent pipeline tailored for systematic reviews. It introduces a novel, expert-knowledge-guided, multi-stage collaborative agent architecture that integrates structured human–agent dialogue, cross-modal information extraction (from text, tables, and figures), and standardized semantic mapping—enabling one-time domain knowledge injection and end-to-end knowledge-driven processing. Contribution/Results: We present the first reusable, end-to-end evidence structuring framework that automatically transforms unstructured scientific papers into unified, machine-readable, standardized evidence tables. Evaluated on a meta-analysis task for NMC811 lithium-ion battery cathode materials, our pipeline reduces analysis time from months to minutes while ensuring high reproducibility. This significantly advances the automation, scalability, and reliability of large-scale literature synthesis.

This work addresses the pervasive issue of citation errors in scientific literature, which existing methods struggle to verify at scale due to reliance on abstracts or limited datasets. We propose BibAgent, an end-to-end agent framework that integrates large language models, cross-source document retrieval, and an adaptive evidence aggregation mechanism, employing tailored strategies for open-access and paywalled publications. A key innovation is the Evidence Committee mechanism, which infers the validity of citations to restricted-content papers through consensus among downstream citing works. To support systematic evaluation, we introduce MisciteBench—a large-scale, cross-disciplinary benchmark comprising 6,350 samples spanning five categories of miscitation. Experiments demonstrate that BibAgent significantly outperforms existing LLM-based baselines in both accuracy and interpretability, enabling efficient, traceable, and scalable detection of citation errors.