물류 안전사고 키워드 분석을 위한 SRoBERTa 및 명사화 기반의 인과 키워드 통합

초록

Recently, many studies have been conducted to extract and analyze causal keywords of logistics safety accidents using natural language processing technology. However, when extracting keywords and analyzing them from large dataset, the complexity of the analysis increases significantly if semantically similar keywords are not integrated. Therefore, this study proposes the causal keyword integration method for the keyword analysis of logistics safety accidents. For this purpose, a dataset was established targeting causal keywords extracted from 3,131 cases of logistics safety accidents, and morphological analysis was performed to nominalize phrases. Furthermore, by applying verb stemming, we executed rule-based nominalization and generated embeddings via the pre-trained language model, SRoBERTa. Through experiments, we confirmed that the proposed method can integrate causal keywords with similar meanings without additional learning processes, by considering the contextual characteristics of causal keywords. Additionally, we verified performance improvements in keyword integration through nominalization.

키워드