Abstract:

Purposes To address the limitations of existing reviewer recommendation methods in semantic matching accuracy and explainability， this paper designs and implements an intelligent reviewer recommendation system （IRRS） integrating a large language model （LLM） with retrieval-augmented generation （RAG）. Methods Using papers published in 2025 in Intelligent Computing as the evaluation dataset， this study constructs an external knowledge base derived from Web of Science literature in computer science and related interdisciplinary fields and develops a dual-layer prompt framework to guide a large language model in generating candidate reviewers with high thematic alignment to the target manuscript. Findings The case-based comparison and evaluator assessments indicate that the proposed system outperforms the Scopus recommendation results in terms of thematic alignment and demonstrates strong explainability in its recommendation rationales. Conclusions The proposed LLM-RAG-based reviewer recommendation framework improves semantic matching and enhances the explainability of recommendation results. By integrating an external knowledge base with a dual-layer prompt design， the framework enables a flexible and scalable recommendation process without requiring modifications to the underlying model architecture. These findings suggest a practical pathway for incorporating large language models into scientific journal peer-review workflows.

Key words: Reviewer recommendation, Large language model, Retrieval-augmented generation, Semantic matching, Explainability, Editorial decision support

摘要：

目的 针对现有审稿人推荐方法在语义匹配精度与结果可解释性方面的不足，设计并实现了一种融合大语言模型（large language model， LLM）与检索增强生成（retrieval-augmented generation， RAG）的审稿人推荐系统（intelligent reviewer recommendation system， IRRS）。 方法 选取Intelligent Computing期刊2025年度已发表论文作为研究样本，构建基于计算机科学及相关交叉领域 WoS（Web of Science）文献的外部知识库，并设计双层提示词框架，引导LLM生成与稿件主题高度匹配的候选审稿人。 结果 案例对比与评审员评估结果表明，系统在主题契合度方面优于 Scopus推荐结果，并在推荐理由可解释性方面表现良好。 结论 基于LLM 与 RAG 融合框架构建的审稿人推荐方法在语义匹配精度与推荐结果可解释性方面均取得积极效果。该框架在不改变底层模型结构的前提下，通过外部知识库与提示词策略的协同设计，实现了可扩展、可配置的推荐机制，为大语言模型在科技期刊审稿人推荐场景中的应用提供了实现路径与方法参考。

关键词: 审稿人推荐, 大语言模型, 检索增强生成, 语义匹配, 可解释性, 编辑决策支持