| Information retrieval and data storage in the information age are very efficient.Entity relationships and attributes in the knowledge graph can be represented by triples,and short texts of entities and related content can be formed into nodes,and relevant nodes can form a structured data system.The query based on knowledge graph can map short entity text to entities in the real world,and query other related information of entities.Previously,knowledge graphs were mainly used in search,question-and-answer and other scenarios.At the same time,researchers are constantly exploring and expanding the usage scenarios of knowledge graphs.The concepts or instances in knowledge graph are unique and uniquely refer to an entity in reality,and its related content can be queried using a unique identifier based on the resource description framework,including the information associated with its entity or the number of occurrences of its concept,etc.The main work done in this paper is as follows:(1)Researched effectiveness of information content based on knowledge graph.This paper studies the knowledge graph data representation model and the related properties of entities,explored and compared the calculation methods of information content in the corpus and the knowledge graph,determines the calculation method of information content based on knowledge graph also has the ability to describe the importance of words.(2)Completed the research on improved semantic similarity based on knowledge graph,and improved the calculation formula of semantic similarity.The concept hierarchy model can calculate the distance between words,and the information content value of words can be calculated based on knowledge graph.Considering the two factors of distance and information content,a new improved similarity calculation formula is proposed,and the similarity is calculated for the semantic similarity test data set.The correlation coefficient calculation is carried out according to the calculation results and the artificial similarity score of the dataset,which proves that the improved similarity calculation method has improved the calculation effect of semantic similarity.(3)Completed the follow-up task comparison experiment based on the improved semantic similarity calculation results.In this paper,two sets of comparative experiments are constructed for word data and document data.Firstly,the types of clustering methods that can directly use the similarity matrix are explored,and on this basis,the text clustering task is completed,and the clustering effect is visualized to enhance the interpretability of the clustering results.It is confirmed by experiments that the improved semantic similarity calculation method based on knowledge graph is suitable for subsequent tasks that rely on similarity calculation,and the experimental data of various clustering algorithms show its feasibility and effectiveness. |
