| With the rapid development of science and technology and the acceleration of technology iteration,the number of patents has also increased explosively.Patents are the most important information carrier of technology and one of the most important research objects of intellectual property analysis.The language of patent documents is accurate,the semantic information of patent documents is complex,and patent documents contain a lot of technical terms,which requires a strong professional background of patent analysts to understand the contents of patents,so most patent analysis work is done manually.This makes it difficult to understand the development of technology quickly and comprehensively.How to apply artificial intelligence technology to patent analysis,effectively obtain the technical information among a large number of patents,generate the resource portrait of intellectual property with knowledge graph,and analyze the evolution and development trend of technology is of great significance.The analysis of intellectual property in this thesis focuses on patents.The main work of this thesis includes the following aspects:(1)This thesis proposes an intellectual property entity recognition method based on transformer and technological word information embedding(BWET).Accurate word vector representation provided by BERT language model is combined with technical word information extracted by IDCNN during the process of word embedding,and transformer encoder with relative position coding is used to improve the representation ability of intellectual property entities and the extraction effect of intellectual property entities.(2)This thesis proposes a construction method of intellectual property resource portrait based on knowledge graph.The patent classification method based on phrase context fusion feature(PCFF)is proposed to make full use of the semantic information of patent text and improve the accuracy of patent text classification.This thesis also provided an intellectual property entity completion method based on text classification,to complete missing key entities of patents and generate the resource portrait of intellectual property.(3)This thesis proposes an evolution analysis method based on intellectual property resource portrait.In order to extract the representation vector of relationships and features between patent nodes and intellectual property entity nodes in the intellectual property resource portrait,an intellectual property resource portrait representation(PKRL)method based on graph convolution is proposed.By Clustering the intellectual property feature vectors in the time window to obtain the research topics of patents and calculating the similarity of topics between time windows to obtain the research topic evolution path through to analyze the evolution of research topics.(4)The resource portrait and evolution system of intellectual property is designed and implemented.The system mainly includes the following modules:intellectual property entity extraction module based on deep learning,intellectual property resource portrait based on knowledge graph,intellectual property evolution law analysis and display module based on intellectual property resource portrait.It mainly implements the functions of intellectual property entity recognition,intellectual property resource portrait construction,retrieval and display of intellectual property evolution. |
