Chinese mineral question and answering system based on knowledge graph

Abstract

Mineral knowledge acquisition is indispensable in geological research. Currently, mineral knowledge can be acquired though a search engine like Google, or though some mineral databases. But usually, the answers from the search engines are not professional enough, and the mineral databases cannot answer the questions input in natural language. In this paper, a Chinese mineral question and answering system based on knowledge graph is proposed with the aim of answering the mineral questions in Chinese natural language professionally, accurately and efficiently. Two deep learning models based on BERT are built to obtain mineral entities and relationships to construct the mineral knowledge graph. After the mineral knowledge graph is constructed, accordingly question templates are defined and another deep learning model based on BERT is built to recognize the intent and the entity/attribute of the input question. The query statement is generated according to the question template pre-defined and is input to the mineral knowledge graph constructed. Then the answer to the question is returned by the knowledge graph. The knowledge graph we constructed has 22,568 entities and 91,699 relationships, which is richer than most of the geological knowledge graph. The question and answering system achieve a 91.2% accuracy on the 2,000 test questions.