NoveltyScope：基于证据的多智能体论文新颖性评估助手
原标题：NoveltyScope: A Multi-Agent Research Assistant for Evidence-Grounded Novelty Assessment of Scientific Papers
AI 摘要
该论文提出 NoveltyScope，一个多智能体科研辅助原型，通过将论文与本地语料库中的先前工作进行检索比对，生成结构化报告和五级新颖性评分。系统提取核心任务与最多两项作者声称的贡献，使用 BM25 与稠密检索及倒数排名融合，并让检察官、辩护人、法官和批评者四个角色基于证据进行评分，法官最多可修订一次。在 50 个 RINoBench 样本上，多智能体评分平均绝对误差低于单 LLM 基线（Gemini 1.26 对 1.48，Qwen3 1.16 对 1.22），向量相似度基线为 1.68。
正文节选
NoveltyScope: A Multi-Agent Research Assistant for Evidence-Grounded Novelty Assessment of Scientific Papers Abstract NoveltyScope is a prototype that compares a scientific paper with retrieved prior work and produces a structured report and a five-level novelty score. The system extracts the core task and at most two author-claimed contributions, then searches a closed local corpus using BM25 and dense retrieval with reciprocal rank fusion. Contribution comparisons request paired passages for p