The Research And Development Of Question Answering System Based On Knowledge Graphs

With the extensive application and continuous development of Internet technology, online information is growing at explosive speed.Hence, the search engine has become an important tool for people to get information online. However, the traditional search engines use string matching rules and always return a large number of related pages. Users need to read these pages to get the answers they truly want. Thus it can be seen that traditional search engines do not understand exactly what users want. Based on this, in this paper, a question answering system based on knowledge graphs that can return the exact answers is studied and proposed.Firstly, the algorithm of knowledge graph representation based on metric spaces is studied in this paper. To overcome the issues of current knowledge graph representation schemes and relation classification, a new relation classification scheme and a new model TransHR which aims at modeling hyper-relational data are proposed. TransHR embeds entities and relations in entity space and relation space, respectively, and transforms the vectors of multiple relations between an entity pair into an individual vector by relation-specific matrices. In the experimental part,two subsets of Freebase are used and the results show that TransHR significantly outperforms current knowledge representation methods.Then, based on the same basic idea, this paper also makes an in-depth study of the question answering (QA) algorithms based on metric space and proposes an effective scheme MSQA. MSQA disassembles the answer triples and introduces the weight matrices. The experimental results show that MSQA has some validity.Finally, a QA system based on the above-mentioned methods is developed, which achieves answering questions exactly. The QA system mainly includes data acquisition module, QA module, background management and front-end display module.