Implicit Information Reasoning And Relation Prediction For Intellectual Property

Scientific and technological intellectual property achievements such as scientific and technological papers and patents are experiencing explosive growth.There is a wealth of valuable information hidden in intellectual property data,and more and more researchers are currently mining intellectual property data.Due to the heterogeneity of intellectual property entities and the complexity and non-duality of their relationships,most current mining algorithms cannot accurately mine information in intellectual property data.It’s crucial to learn how to efficiently extract implicit information from large-scale intellectual property using artificial intelligence technology.Hypergraph modeling and hypergraph learning are used to mine intellectual property data,enabling reasoning about implicit information in intellectual property,predicting relationships between concepts,and locating potential collaborative relationships between intellectual property authors.The main work accomplished in this thesis is as follows:(1)This thesis proposes a hypergraph structure representation learning method for intellectual property.By using hypergraph line expansion convolution channels and hypergraph convolution channels to extract hypergraph structure information,and using adaptive fusion methods to fuse the features of the two channels,it achieves the representation of non-binary relationships and heterogeneous entities in intellectual property data.Experimental results show that the proposed hypergraph learning method can better characterize the hypergraph structure information of intellectual property rights,avoiding the problem of hypergraph structure loss.(2)This thesis proposes a hypergraph semantic learning method based on feature similarity attention.In order to obtain feature similarity attention,an intellectual property keyword extraction layer based on title features is designed to extract keywords and embed text.An attention coefficient matrix is constructed using text feature similarity between intellectual property entities.The information aggregation functions of hyperedges and hypernodes in intellectual property hypergraphs are designed,and the information of hypernodes and hyperedges is aggregated,respectively,to achieve semantic learning of intellectual property hypergraphs.Experimental results show that the proposed method can efficiently extract keywords from intellectual property text and obtain high-quality vector representations of intellectual property.(3)This thesis proposes an implicit information reasoning and relationship prediction method for intellectual property.The method uses hypergraph width first traversal to obtain hyperpaths,implicit information reasoning is implemented for intellectual property author entities that exist between paths,and potential cooperative relationships between these author entities are mined.The semantic similarity between author entities is calculated,which serves as a basis for evaluating and mining potential cooperative relationships.The experimental results show that the proposed method can effectively accomplish implicit information reasoning and relationship prediction for intellectual property data,mining out potential cooperative information.(4)An intellectual property data implicit information reasoning and relationship prediction system is designed and implemented,which mainly includes the following modules:intellectual property data increment and hypergraph calculation module,intellectual property author entity retrieval module,and intellectual property author entity information display module.Functions such as hypergraph modeling of intellectual property data,learning and representation of intellectual property hypergraphs,prediction of collaborative relationships between intellectual property authors,author entity retrieval,and intellectual property author information display are implemented.The system is tested,and the test results show the effectiveness and stability of the system.