Knowledge Base Question Answering System Based on Knowledge Graph Representation Learning
Yue Wang, Qimai Chen, Chaobo He, Hai Liu, Xiyu Wu
Abstract
Knowledge Base Question Answering (KBQA) refers that questions are answered by acquiring the relationship or entity from knowledge graph. The knowledge base is being high frequent used in modern question answering systems, which can find the exact answer from the knowledge base and return it to the users. However, due to the flexibility, richness and fuzziness of natural language, it is not easy to match the semantic information of questions with the text answers. How to accurately match these natural language questions with a large number of knowledge graph and improve the accuracy of questions and answers is an urgent problem to be solved. In this paper, we propose a question answering algorithm named TransE-QA based on knowledge graph representation learning to solve Simple QA problem. This algorithm is an end-to-end algorithm for question answering of simple QA problem. TransE and TextCNN are mainly used to represent the knowledge graph and the question. Based on knowledge graph representation learning, we propose a new scoring function, which the answer can be returned directly that required by the question. Besides, based on the TransE-QA algorithm proposed in this paper, we develop a KBQA system to visualize the process. The experiment shows that the algorithm TransE-QA, which proposed in the thesis, achieves 80.2% accuracy on FB5M dataset. It achieves better performance than previous state-of-the-art methods.