Застосування LLM в інформаційній системі для тематичного аналізу користувацьких коментарів

Abstract

This work presents a prototype of information system for LLM-powered thematic analysis of user comments. The second phase of thematic analysis – codebook-based coding – is implemented as an auditable sequence of five operations with stored input, intermediate and output artifacts. A PoC was run on the set of 50 Reddit comments using two models (GPT-4o-mini, GPT-5.2). Results were compared with a researcher-created golden coded set. The paper acknowledges the limitations of a single-coder gold standard and outlines next steps, including involving independent coders and scaling to larger datasets