Keybert와 Bertopic을 활용한 텍스트마이닝 연구동향 분석

초록

Text mining techniques for analyzing and utilizing large-scale text data are widely used not only in engineering but also in social sciences, education, and almost all academic fields. This study aims to collect and analyze domestic academic and thesis papers on text mining using the latest text mining techniques. For this purpose, papers were collected from the Research Information Sharing Service (RISS) database using the keyword ‘text mining’, and keyword analysis and topic modeling were conducted on the collected papers. In keyword analysis, frequency-based analysis using TF-IDF and analysis using BERT-based KeyBERT were compared. Additionally, in topic modeling analysis, traditional statistical-based LDA techniques were compared with BERTopic, a topic modeling technique based on the latest BERT language model. The results showed that BERT-based topic analysis demonstrated superior performance in terms of coherence score. Particularly, the topic extraction based on Korean embedding models and Keybert recorded higher coherence scores compared to those based on multilingual models and sentence-based extraction. Through these findings, this study aims to present the applicability and potential of the latest techniques in Korean text mining.

키워드