Research Of Open-world Knowledge Completion Technology For Text Question And Answer

Problems such as sparse data and incomplete knowledge restrict the development of knowledge graph.In order to improve it there have emerged lots of researches on knowledge completion algorithms.The traditional knowledge completion algorithms can complete the knowledge graph to some extent,but it can not make effective use of the knowledge in the knowledge graph.And because the traditional knowledge completion algorithms ignored the open world and the time information contained in knowledge,the knowledge graph used in question answering system is difficult to provide knowledge support correctly and effectively.To overcome these problems,this thesis carries out the research on the open world knowledge completion technology for text-oriented Q & A,and implements two related models respectively.Through the construction of the OpenWorld Knowledge Dynamic Fusion model,the open-world knowledge is incorporated to complete the knowledge graph and improve the effect of knowledge representation.The Knowledge Evolution based on Time Series Change model is proposed to incorporate time information into the knowledge completion model and and constructs time hyperplane to improve the completion effect on temporal knowledge graph.The main work of this thesis is as follows:(1)A dynamic knowledge fusion model for open world is proposed and realized.Most of the current knowledge completion models have the problems of insufficient utilization of knowledge graph information and neglect of open world knowledge.In order to solve this problem,the Open-World Knowledge Dynamic Fusion model is proposed in this thesis.The model improves the traditional knowledge completion model DKRL,adopts a text description embedding model that combines attention mechanism and convolution neural network when obtaining entity and relational semantic vectors.And the model integrates open world knowledge,which is used to assist the task of knowledge completion and expands its scope of application.The experimental results show that compared with the baseline models such as Tran E,Trans H,DKRL and OWE,the average ranking and hits@10 index of this model are the best in the entity prediction experiment.(2)In this thesis,a knowledge evolution model based on time series change is proposed and implemented.The existing traditional knowledge completion models seldom consider time information,so it is difficult to cope with the task of time knowledge graph completion.On the other hand,the dynamic knowledge graph completion model oriented to time knowledge graph is limited to the structure information of knowledge graph when obtaining knowledge representation vector.In order to solve this problem,The Knowledge Evolution based on Time Series Change model is proposed in this thesis.The knowledge representation process of the model in the knowledge completion model integrates time information,and combines the open world knowledge mentioned earlier to assist in the task of time knowledge graph completion.The experimental results show that compared with the baseline models such as Trans E,Trans H and Hy TE,the average ranking of this model is the best compared with hits@10 in the entity prediction experiment on the time knowledge graph,and the head entity prediction and tail entity prediction of the optimal comparison model Hy TE in the hits@10 index are improved by2 and 8 thousand points respectively.(3)Based on the above research results,this thesis proposes and constructs a text question answering system based on open world knowledge completion technology.The system consists of three functional modules: question processing,data retrieval and question answering,which has the advantages of high accuracy,strong interpretability and visual display of question answers.……...