Research On Patent Knowledge Search Based On Knowledge Graph

Patent,as the knowledge carrier that can best reflect human's ability of invention and creation,also plays a crucial role in the development of manufacturing enterprises.On the one hand,enterprises can protect their intellectual property rights by applying for patents;on the other hand,enterprises can promote their own innovation by rational utilization of the knowledge contained in patents.With the rapid development of artificial intelligence technology and the increasing demand of China's manufacturing enterprises for intelligent management of their knowledge,the combination of the two situation will have a broad application prospect for effective knowledge management and mining analysis.At present,knowledge management in manufacturing enterprises has not applied cutting-edge deep learning technologies such as natural language processing to the scene of knowledge mining and analysis,which leads to the problem of low knowledge utilization efficiency.So this paper attempts to apply knowledge graph technology to the search of typical knowledge owned by manufacturing enterprises--patent,and carries out the following research work.In order to refine the patent from the document level to a smaller structured knowledge granularity so as to realize more refined knowledge management and application,the research of patent knowledge graph construction will be carried out.In this paper,typical engine-related patent texts possessed by machinery manufacturing enterprises will be used as data sources to define the ontology layer of the patent knowledge graph.Then we'll research the automatic extraction of entities and relationships between entities in the patent text with deep learning algorithm based on Bert model,using objectively labeled patent data set prove the accuracy of the patent knowledge extraction.Finally,the patent knowledge graph is going to be constructed by the extracted knowledge of correct entities and relationships to realize the structural modeling of patents.In order to carry out the computable process on constructed patent knowledge graph and support its digital application efficiency,we'll research the vectorization representation technology of the knowledge graph.In this section,feasibility of the existing TransE algorithm will be analyzed and some targeted improvements will be made.we'll propose a WTBT algorithm that integrate lexical information and entity's type information.Link prediction task will be used to verify the general effectiveness of the proposed algorithm and ensure that the general evaluation metrics will be improved.By using more intuitive methods such as vector distance calculation,vector visualization,similar term entities in patent knowledge graph will be found,and the vector obtained by our algorithm will be finally verified to have effective semantic expression ability for term vocabulary.In order to improve issues that the existing patent search system's single mode in text matching is unable to expend the input retrieve statement organically and appropriately which results in limiting the patent reuse efficiency,we'll classify patterns of search statement,analysis different intentions of specific pattern,and propose Patent Extend search model Based On Knowledge Graph(PEBKG).Our PEBKG model is going to present different extending search methods for different search demands and we'll take some search instance to prove the validity of the model.The research content of this paper provides a new idea for the knowledge mining method of patent text,and uses the graphical structuring method of knowledge graph to deconstruct patent text,supports the intelligent search requirements with associative characteristics,so it's innovative in some way.There is practical and referential value in the construction and application of patent knowledge graph.