Research On Patent Automatic Recommendation Technology Based On Knowledge Graph

With the vigorous implementation of innovative development strategy in China,patent,as an intellectual property with high scientific and technological content,is increasingly being valued by enterprises and users.In the face of massive amounts of patent data,mastering the scientific and technological knowledge contained in patents is of great significance to promote the development of enterprises and the country.In the process of patent data analysis,how to divide the technic al fields of patents conveniently and efficiently is a problem worthy studying.This paper introduces the idea of recommendation system,combined with actual demand,and recommends patents according to the similarity between patent text data,so as to realize the recommendation of unknown patent technology fields.Since the knowledge graph is widely used in the recommendation field,this paper mainly studies the related technology of patent knowledge graph construction,the vectorized representation of patent related entities,and the technical field recommendation based on the patent knowledge graph.The main research contents are as follows:Firstly,this paper studies the related technologies of constructing patent knowledge graph,analyzes the patent title,abstract,applicant and other data,determines patent-related entities and relationships,and completes the construction of the patent domain ontology database.Then,the entity,attribute and relational data contained in the patent text are extracted,and the processed data are stored in the Neo4 j graph database to complete the construction of the patent knowledge graph.Secondly,the entity is vectorized according to the Trans E model.In the process of model training,the original negative sampling algorithm is improved and the patent text entity information is embedded in a 100-dimensional vector.The vectorized patent entity contains the semantic relationship between the entities in the patent knowledge graph,which paves the way for the realization of patent recommendation.Finally,this paper proposes a patent recommendation algorithm based on the knowledge graph(KG-PR),which integrates the vectorized result of the patent knowledge graph into the content-based recommendation algorithm.The similarity between patents is calculated by using patent title entities in the knowledge graph to realize the recommendation of similar patents,so as to complete the recommendation of the technical field of the patent.In order to verify the rationality of the method in this paper,the technology fields of 100 patents are recommended by comparing with the content-based patent recommendation algorithm(CB-PR).The predicted technical fields are compared with the actual ones,and the proportion of the correct prediction is calculated.The experimental results show that the KG-PR algorithm proposed in this paper has higher correct rate than the patent content-based recommendation algorithm.In the end,according to the IPC number group,100 patent technical fields can be recommended.It can be concluded that the correct rate of the patent recommendation algorithm based on the knowledge graph in recommending one IPC number is 82%,and the correct rate of recommending multiple IPC numbers is 98%,which shows the feasibility of the KG-PR algorithm proposed in the paper in patent recommendation.