Document Classification System Using BERT

Abstract

Research is actively being conducted to create meaningful value using big data generated across society. Accordingly, domestic and international papers, patents, e-books, etc., have been databased and are being used for various studies on research and technology trends. This paper has designed and developed a document classification system based on BERT. For experimental performance evaluation, abstracts from the four most prestigious conferences in the field of information security were collected and used to compare a document classification system using BERT with one based on SBERT. The classification results showed that the BERT-based document classification system was about 11.7% superior, and this paper presents ways to develop a better document classifier.