Intelligent Audit Question Answering System based on Knowledge Graph and Semantic Similarity

Abstract

Audit work needs to refer to a large number of law and institutional documents, and the business process is intricate. Finding the answer to audit question based on the knowledge graph can improve the work efficiency observably. This paper has proposed a knowledge base for the electric power audit task. In detail, this project collects law texts related to auditing, and has designed schema and specific rules for them to extract triples. These triples are stored in the Neo4j graph database as Knowledge Graph. In addition, this paper makes templates to generate a question answering dataset about electric power audits. According to the audit knowledge graph, a character-level convolutional neural network is trained and an intelligent audit question answering system is built based on the semantic similarity. The effectiveness of the system is evaluated by experiments.