GPT-3による日本語テキストからの因果関係抽出のためのWikidata中の因果関係知識を用いた学習データ拡張の検討

抄録

Causal relation knowledge is necessary to develop a facilitator agent that can understand discussion points and participants' opinions. However, it is not enough to be included in the Knowledge Graph. In this study, we attempted to extend the training data using Wikidata's casual relation knowledge as a method for extracting causes. To compare whether the proposed extraction method is more accurate than previous methods, we compared the accuracy of the output causes by inputting sentences. In addition, a calculation method was examined to determine if the extracted causes could be considered a general causal relationship. As a result, the accuracy of the extraction is improved over conventional methods, and a threshold value can be determined to consider it as a general causal relation. Future work includes the development of a facilitator agent to support discussions using the methods in this paper.