Research On The Construction And Application Of Innovation-oriented Patent Knowledge Graph

Product innovation is function innovation in a certain sense.Mining the innovation knowledge in patents is conducive to technical evasion and breaking patent barriers.Knowledge graph is an efficient knowledge expression model,which can effectively find the connections between entities in the graph,and provide support for upper-level applications such as intelligent search and knowledge answering.Therefore,on the basis of patent analysis,mining effective patent knowledge and constructing an innovation-oriented patent knowledge graph have a very important role.Extracting innovative knowledge from patents is the focus of patent analysis research.At present,patent analysis is not very good in feature extraction.There are problems such as incomplete feature extraction and unreasonable extraction,and the relationship between patents cannot be well explored.Building an innovation-oriented patent knowledge graph requires not only a good ontology architecture,but also rich data filling and relationship building.In response to the above issues,the main contributions of this article include:(1)The key technologies of the patent knowledge graph for innovation are studied.First,the top-down approach is used to construct the related ontology of the patent knowledge graph for innovation.Taking the construction of patent knowledge graph in the field of chemistry as an example,the related technical schemes from knowledge extraction,knowledge discovery,and knowledge storage were designed and implemented.In order to improve the recall rate and cover more patent knowledge,based on Wikipedia,Synonym,etc.,the expansion of the patent knowledge graph metadata for innovation was carried out,and the expansion words were linked to the knowledge base through physical links.(2)In the knowledge discovery stage,a dynamic clustering hybrid model is proposed.This model combined deep learning-related technologies to extract functional information features from patents that express similar functions.The model feature extraction was mainly divided into deep semantics representation section and functional word attention section.The deep semantic representation part used the bidirectional long short-term memory network combined attention mechanism to extract the text sequence features,used the convolutional neural network to extract the text embedding features,extracted the text topic features with the improved weight Latent Dirichlet Allocation,and added functional words to the functional word focus part to strengthen the role of these functional words in clustering.During the clustering process,this study adjusted network parameters,completed the clustering of patents,and mined the relationship between different patents.Later,in the process of extracting innovation knowledge,the topic extraction model was used in combination with TF-IDF and Text Rank algorithms to extract innovation patent knowledge.(3)The graph database Neo4 j is used to store the organized structured triples knowledge.At the same time,the system structure of the innovative patent knowledge graph was designed,and a platform for knowledge patent graph for innovation patents was constructed.The platform has patent search and ontology architecture display functions,provides a friendly visual interface and API services,and integrates the proposed algorithm into a complete application.In this paper,the existing patent data in the research institute and the purchased patent texts were used as experimental data.Under the same experimental conditions,relevant comparative experiments were performed for the dynamic clustering hybrid model.The F-measure of this method is 87.954%,which is the highest in comparison experiments and verifies the validity of the model.