Research And Application Of Image Content Understanding And Expression Method Based On Deep Learning

The intelligent question answering system based on knowledge graph is supported by structured knowledge,realizes the accurate output of data to information and then to knowledge,and significantly improves the efficiency of information acquisition.In real scenarios,knowledge is time-sensitive,and there are common problems such as lack of semantics.For this reason,it is necessary to consider the impact of time on entity states and relationships,so as to realize the construction of dynamic knowledge graphs.On this basis,it is necessary to mine the implicit associations between entities in the dynamic knowledge graph,capture the dynamic evolution process of knowledge,and realize the completion of the missing semantic information in the dynamic knowledge graph.For this reason,this paper takes the financial intelligent question answering system based on knowledge dynamic completion as the research topic,and conducts research on the problems and challenges of dynamic knowledge graph completion technology.The global semantic information of subgraphs improves the accuracy and robustness of dynamic knowledge graph embedding,and realizes accurate completion of knowledge graphs.In addition,for scenarios in the financial field,an intelligent question answering framework based on knowledge graph is constructed,which combines two question answering methods based on rule matching and dynamic knowledge graph representation to realize adaptive answers to time-series and non-sequential questions.The main work and contributions of this paper are as follows:（1）A dynamic knowledge graph completion method based on global semantic capture of timing subgraphs is proposed.By capturing the global semantic information of each timing subgraph Gt and fusing it with the local information of each node in the subgraph,the accuracy of the timing subgraph embedding are improved,which fully express rich semantic information contained in the dynamic knowledge graph,so as to optimize the quality of the dynamic knowledge graph completion.The experimental results show that the dynamic knowledge graph completion model proposed in this thesis has improved Hits@3 and Hits@10 scores on multiple public datasets.（2）Based on the above research results,a time series knowledge graph question answering system for the financial field is designed and constructed.The system adaptively selects the method based on rule matching and dynamic knowledge graph representation to realize intelligent question answering of time series questions and non-sequence questions in the financial field.Taking the massive Internet data in the financial field as the data source,defining the entity types and relationships,taking the structured knowledge as the knowledge support of the question answer system,and each functional module of the system is realized.The construction of the intelligent question answering system helps people to easily obtain the financial information they need,and has strong application value.