迈向自动化研究：利用具有范畴结构的论文知识图谱挖掘可证伪的研究构想
文章背景与核心概要
当前的自动化科研构想生成系统大都构建在大语言模型（LLM）基础之上，但它们存在一个根本性的结构缺陷：系统将学术论文视为扁平的对象（字符串、向量或随机配对），从而丢弃了研究人员在进行跨领域类比推理时自然依赖的有类型关系——即问题、方法、指标和断言。
本文引入了一个范畴论框架，将单篇论文建模为小型范畴，把跨论文的构想生成转化为偏函子保持（partial functor preservation）。该方法实现为一个三层算法（范畴签名聚类、函子保持门控以及六轴LLM合理性评判器），在保持高比例定量可证伪性的同时，高效过滤了跨领域候选构想。
目录 / 摘要 (Summary)
Current automated research-idea generation systems built on Large Language Models (LLMs) suffer from a fundamental structural flaw: they treat academic papers as flat objects (strings, vectors, or random pairings), discarding the typed relationships—problems, methods, metrics, and claims—that researchers naturally rely on for cross-domain analogical reasoning.
This paper introduces a categorical framework that models individual papers as small categories, turning cross-paper ideation into partial functor preservation. Implemented as a three-layer algorithm (categorical signature clustering, a functor-preservation gate, and a six-axis LLM plausibility judge), the approach efficiently filters cross-domain candidates while maintaining a high rate of quantitative falsifiability.
当前构建在大语言模型（LLM）之上的自动化研究构想生成系统存在一个根本性的结构缺陷：它们将学术论文视为扁平的对象（字符串、向量或随机配对），从而抛弃了研究人员在进行跨领域类比推理时自然依赖的有类型关系——即问题、方法、指标和断言。
本文引入了一个范畴论框架，将单篇论文建模为小型范畴，从而将跨论文的构想生成转化为偏函子保持。该方法实现为一个三层算法（范畴签名聚类、函子保持门控以及六轴LLM合理性评判器），在保持极高定量可证伪性的同时，高效地过滤了跨领域的候选构想。
元数据 (Metadata)
- arXiv ID: 2608.20361
- Authors: Yuchen Wang, Zhongzhi Luan
- Subjects: Computation and Language (
cs.CL); Artificial Intelligence (cs.AI)- ACM Classes: I.2.7; H.3.3
- Submitted on: June 17, 2026
- Comments: 18 pages, 10 figures
- arXiv ID: 2608.20361
- 作者: Yuchen Wang, Zhongzhi Luan
- 学科分类: 计算与语言 (
cs.CL); 人工智能 (cs.AI) - ACM 类别: I.2.7; H.3.3
- 提交时间: 2026年6月17日
- 评论: 18页，10张图表
摘要 (Abstract)
Automated research-idea generation systems built on large language models (LLMs) share a structural weakness: they reduce ideation to free-text recombination, random paper pairing, or embedding-similarity retrieval. The three approaches fail in the same way: each treats a paper as a flat object, a string or a vector, and so quotients away the typed problem-method-metric-claim arrows a researcher actually uses when reasoning about a cross-domain analogy.
We recover the missing structure with the minimal piece of category theory that a typed graph alone does not provide: composition, together with identity arrows, which makes it possible to ask whether a proposed analogy preserves relation chains.
Concretely: 1. Each paper \(p\) is modelled as a small category \(C_p\) whose objects are extracted typed research entities and whose morphisms are the relations the paper asserts. 2. A cross-paper bridge from \(p\) to \(q\) is defined as a partial functor candidate \(F: C_p \rightarrow C_q\) that preserves object kinds and covered relation classes. 3. We instantiate the model as a three-layer algorithm: * Categorical signature clustering * A functor-preservation gate * A six-axis LLM plausibility judge
构建在大语言模型（LLM）之上的自动化研究构想生成系统共享着一个结构性弱点：它们将构想生成简化为了自由文本重组、随机论文配对或嵌入相似度检索。这三种方法在根本上如出一辙：它们都将论文视为扁平的对象（字符串或向量），从而将研究人员在进行跨领域类比推理时实际使用的有类型“问题-方法-指标-断言”箭头给抽象掉了。
我们利用范畴论中单纯靠有类型图谱无法提供的最小核心组件来恢复这种丢失的结构：复合（composition）以及恒等箭头（identity arrows），这使得我们能够判断一个提出的类比是否保持了关系链。
具体而言： 1. 每篇论文 \(p\) 被建模为一个小型范畴 \(C_p\)，其对象为提取出的有类型研究实体，其态射（morphisms）为论文所断言的关系。 2. 从 \(p\) 到 \(q\) 的跨论文桥梁被定义为一个偏函子候选 \(F: C_p \rightarrow C_q\)，它保持了对象的类别以及所覆盖的关系类。 3. 我们将该模型实例化为一个三层算法： * 范畴签名聚类 * 函子保持门控 * 六轴LLM合理性评判器
Evaluated on a corpus of tens of thousands of full-text-parsed papers under four ablation conditions, the categorical gate filters cross-domain candidates at roughly a 17:1 ratio while the quantitative-falsifier rate of accepted ideas stays above 83% throughout. Furthermore, every rejected candidate is retained with its per-axis rationale, allowing the gate to double as a logging layer rather than a silent filter.
在包含数万篇全解析论文的语料库中，并在四种消融条件下进行了评估，结果表明：范畴门控以大约 17:1 的比例过滤了跨领域候选，而被接受构想的定量可证伪率始终保持在 83% 以上。此外，每一个被拒绝的候选都连同其分轴理由一并保留，使该门控能够充当日志记录层，而不仅仅是一个静默过滤器。