Automated comparison of scientific research based on academic genealogy

The article addresses the problem of analyzing large corpora of scientific texts. It is noted that traditional scientometric approaches based on citation analysis do not always capture the substantive proximity of ideas or trace the intellectual continuity of scientific knowledge. The paper proposes an algorithm that combines the capabilities of large language models (LLMs) with quantitative analytical methods. The approach proceeds in two stages. At the first stage, LLM is used to automatically structure the text of a scientific paper and convert it into a machine-readable format. This yields, for each work, a vector representation of its thematic profile. At the second stage, thematic profiles of individual works and of pre-specified groups (schools of thought) are compared using centroid distances and clustering techniques. The ultimate goal is to move beyond formal relational analyses among researchers (e. g., “supervisor — dissertation candidate”) toward constructing an intellectual genealogy that tracks the evolution of schools of thought on the basis of the substantive proximity of their ideas. It is emphasized that the proposed approach does not replace expert assessment but serves as an auxiliary tool for navigating the scientific literature and identifying connections between studies.