Auto-growing knowledge graph-based intelligent chatbot using BERT

Abstract

It is very important to get computers to understand human common sense. A knowledge graph that links words based on relationships is an important skill that allows computers to learn common sense easily. However, knowledge graphs, devised by many existing studies, consist only of specific languages or fields and have limitations that cannot treat neologisms. In this paper, we propose a chatbot system that collects and analyzes data in real-time to build an automatically scalable knowledge graph and utilizes it as the base data. In particular, the fine-tuned BERT-based model for relation extraction is to be applied to auto-growing graphs to improve performance. By building a chatbot where human common sense is learned by using auto-growing knowledge graph, it verifies the availability and performance of knowledge graph. ICIC International © 2020.