Development of a terminology-aware text similarity evaluation method using technical term extraction

Abstract

In this study， we extract technical terms from text using GiNZA， a Python-based natural language processing library， and NPYLM， an unsupervised language model capable of learning word segmentation． Sentences are then vectorized using λ-fuzzy measures， and their similarity is calculated using the Choquet integral．Our method enables effective extraction of domain-specific terminology， which is difficult with conventional morphological analysis． By incorporating these terms into similarity computation， we propose a more accurate similarity metric compared to existing methods．