Research And Application Of Question Answering System Based On Knowledge Graph

The rapid development of artificial intelligence technology has spawned a series of new technologies,and the automatic question answering system is the representative achievement.Unlike traditional search engines,the question answering system can answer fact-type questions with accurate and brief answers,reducing user usage costs.The knowledge graph is a new form of knowledge database,and the data logically constitutes a huge network.Compared with the traditional relational database,the knowledge graph can better describe the semantic relationship between entities,which is in line with human understanding of the objective world.The question answering system based on the knowledge graph combines the advantages of the automatic question answering system and the knowledge graph,and has attracted more and more attention.This paper studies the key technology and application of knowledge graph based question answering system.This paper first introduces and studies the key technologies,including entity recognition,entity link and relationship detection.Entity recognition and entity linking are the pre-tasks of the question answering system and play a key role in the final question answering effect.This article introduces the definition of these two tasks,common methods and evaluation indicators.For the answers to simple questions,this article introduces the relationship detection model based on residual connection in detail,and improves it with the latest pre-training model technology.In response to complex questions,this paper proposes a semantic analysis method based on semantic graphs.The semantic graph is the intermediate representation of natural language questions and SPQARQL logical expressions.Through pre-defined actions,all possible semantic graphs can be generated without hesitation.The logical structure of the semantic graph is a directed graph.For this feature,this paper uses a gated neural graph network to encode the semantic graph,and on this basis,a multi-task learning method is tried to further improve the model capabilities.The experimental results show the effectiveness of our proposed method.Although various methods have been proposed for the knowledge graph based question answering system,in practice,they have encountered a dilemma.The reason for this dilemma is mainly due to the mismatch between knowledge graph and training data.In response to this phenomenon,this paper proposes a method based on question generation.This method expands the triple into a completed question through templates.When the user makes a query,the system uses the full-text search technology and the semantic matching model to retrieve the pre-generated question with the highest matching degree,and returns the corresponding triple to the user.The method based on question generation proposed in this paper can develop a usable knowledge graph question answering system with lower labor cost.On the occasion of the centennial celebration of Harbin Institute of Technology,this paper constructs a knowledge graph with Harbin Institute of Technology campus information as the theme,and marks the data for this knowledge graph.This knowledge graph is the first knowledge base with the theme of university information at home and abroad.On this graph,this paper compares and analyzes the effects of the two solutions of question answering,and develops an available question answering system.