Research Of Domain-knowledge-based Restricted Domain Chinese QA System

The expressing, storing and sharing of the domain knowledge are the main work of enterprise knowledge management. At present, the knowledge retrieval in knowledge management system is usasully implemented by keyword matching. The keyword matching method has a limited precision of knowledge retrival because it can not understand the users'requirement completely and can not take full advantages of the domain knowledge.For this issue, according to the knowledge retrieval requirement of an aircraft engineering institute, this thesis first analyzes the enterprise domain knowledge and gives a domain knowledge framework. Based on the domain knowledge, a semantic character based question taxonomy method is proposed, and a multi-strategy answer retrieval framework is designed. The framework is combined of three knowledge retrieval method: the dependency and semantic similarity question matching method, the structural query based repository retrival method and the TRIZ contradiction matrix based interaction retrieval method. According to the results of experiment, the system is qualified for the knowledge retrival of the enterprise.This thesis'main work and characters are followed:1. Analyzing the consisting framework of enterprise knowledge. The framework in this thesis contains the enterprise repository, domain thesaurus, erperts QAs and the TRIZ contradiciton matrix.2. A charaters matching based question taxonomy method is proposed. The method retrieves the question's charaters by natrual language processing, and establishs the question category by pattern matching.3. The thesis proposes a semantic information and dependency relationship based question similarity method. The method analyses the sentence semantic information and dependency relationship to calculate the similarity of questions, and effectively increase the precision of question matching.4. An interaction innovation principle query method is proposed. This method is based on TRIZ contradiciton matrix. It selects the technical contradictions by the feedback of the QA system users, and provides the users innovation design principle.