Research On Knowleage Base Question Answering Techniques For Multi-turn Scenarios

With the continuous updating of information technologies such as the Internet,the Internet of Things,and cloud computing,the amount of data is exploding.How to accurately obtain the required information has become a focus issue in both academia and industry.Although search engines can return a large number of relevant web pages,users often need to spend a lot of time and energy to find accurate answers.Knowledge graph-based question answering aims to directly provide accurate answers based on users’ natural language questions.Currently,knowledge graphbased question answering technology has achieved good results on single questions,but its ability to obtain contextual information in multi-turn question-answering scenarios is poor,resulting in a decrease in accuracy.In addition,because knowledge graphs rely on manual construction,compared to massive unstructured text corpora,there are issues with missing entities and relationships in the internal information of knowledge graphs,which can affect the accuracy of question answering.Therefore,this paper studies the historical information tracking technology and knowledge graph and unstructured text fusion question answering technology in knowledge graph-based multi-turn question answering.For the problem of historical information tracking,this thesis proposes a multi-turn question-answering method for knowledge graph based on historical information.Specifically,in the entity tracking process,a joint encoding method of question information and entity information is proposed,and in the answer prediction process,a historical information tracking method based on attention mechanism is proposed.Experimental results on real datasets show that the proposed historical information tracking method for multi-turn question answering improves the accuracy compared to existing methods.For the problem of knowledge graph and text fusion,this thesis proposes an answer retrieval method based on graph neural networks.Specifically,a subgraph generation algorithm based on seed transfer graphs is proposed to apply the knowledge graph and text fusion question answering algorithm to multi-turn question answering scenarios,and a subgraph extension algorithm based on entity connections is proposed to supplement missing information in the knowledge graph with text information.Experimental results on real datasets show that the proposed knowledge graph and text fusion question answering method improves the accuracy of question answering in multi-turn question answering scenarios compared to existing methods.