A Knowledge Graph Completion Method Fused With Adaptive Enhanced Semantic Information And Application

Knowledge graph is a technology that uses semantic relationships to represent and organize knowledge.It builds a huge semantic network by integrating massive knowledge resources in a networked way so that data can be organized and understood in a way similar to human cognition,thus providing more accurate services to people.The construction of the knowledge graph is a complex process,one of the most important parts of which is the complementation of the knowledge graph.When constructing a knowledge graph,since a large amount of information comes from literature and the web,difficulties may be encountered in extracting this information,resulting in a lack of completeness or accuracy in the knowledge graph.The purpose of knowledge graph complementation is to make the knowledge graph more complete and accurate by inferring the missing information.Trans E model and Trans H model are the more common knowledge graph complementation models at present.They are both based on the idea of translation to perform knowledge graph complementation,but they tend to ignore the rich semantic information in the triad in the process of making inference,so there are certain limitations.To remedy this shortcoming,this paper constructs a knowledge graph complementation method that incorporates adaptively enhanced semantic information,and explores related applications to design and implement a movie Q&A system based on knowledge graph complementation.The main work is as follows:(1)In order to obtain the hidden semantic information inherent in the triad,this paper inputs the textual representations of entities and relations in the triad into the BERT model for fine-tuning to capture the semantic information of entities and relations,thus generating a high-dimensional triadic word vector with semantic information.(2)Considering that high-dimensional vectors can have an impact on the experimental results,this paper refers to the BERT-whitening model and uses the method of dimensionality reduction of high-dimensional word vectors,which is achieved by the principle of co-variance and orthogonal transformation,so as to generate more efficient semantic vector representations.(3)In this paper,we also pay attention to the semantic information attention representation of entities and relations in triads,and adopt the method of attention feature embedding to calculate the semantic attention score between relations and entities in positive and negative triads,and fuse it into the structural information to form the soft constraint rule of semantic information,which is added to the Trans H translation model to realize the adaptive enhancement of semantic information.(4)Through experimental comparison on FB15 K and WIN18 datasets,the method proposed in this paper has a certain improvement in effect compared with the original translation model,which verifies the rationality and effectiveness of the method,thus enhancing the expression effect of the translation model to a certain extent.On the basis of completing the research work on key issues,this paper also carries out the overall design and implementation of the movie knowledge quiz system,and conducts relevant tests and analysis on the system to prove the usability,ease of use and user-friendliness of the system.