Research On Key Technologies Of Automatic Patent Recommendation Based On Citation Network

The report of the Nineteenth National Congress emphasized the need to firmly implement the innovation-driven development strategy and accelerate the construction of an innovative country.As a patent document that integrates economics,legality and technicality,it is one of the important supports for achieving innovation-driven strategies.In addition,patents It is the carrier of advanced scientific and technological information,so the analysis of patent information is the first boost for the country and enterprises to implement innovation-driven development,which is conducive to the continuous improvement of their "software" strength and innovation capabilities,as well as industrial layout and technical management..However,with the emergence of massive patent data,how to quickly identify core technologies and emerging technologies is a problem that countries and enterprises need to solve urgently.Based on the citation network,this paper conducts research on key technologies for automatic patent recommendation from core technology field identification,core patent identification,and emerging field discovery.First,divide the patents according to the IPC classification number to obtain the directed connection graph between the technical fields,and then calculate the sum of the citations of the patents contained in each field as an authority factor matrix,and at the same time calculate each field in the citation network The intermediary centrality constructs the intermediary factor matrix,and finally brings these two matrices into the Page Rank algorithm to obtain the PBR algorithm,which is a recognition algorithm in the core technical field.Secondly,based on the Page Rank algorithm,this paper considers the connection attributes(cited times,patent age)and own attributes(number of patents of the same family,number of claims)of patents,and introduces the ratio of in-degree factor,time factor,patent factor of the same family and rights factor,An improved Page Rank algorithm or PPR algorithm is proposed.Finally,this paper converts the directed citation network to construct an undirected weighted co-citation network.Then,based on the LFM algorithm,it combines the redundancy detection function and the undirected PRcen algorithm,as well as the "isolated nodes" according to the degree of attribution.The distribution processing method finally obtained the discovery algorithm in the emerging field,namely the PLFM algorithm.In order to verify the rationality and superiority of the proposed algorithm,this paper selects the patent data of aviation,spacecraft and equipment manufacturing industry for empirical analysis.First,by comparing the ranking of the results of the PBR algorithm and the PR algorithm,it is found that the two algorithms havetheoretical consistency.In addition,compared with the PR algorithm,the PBR algorithm can better identify the core technology areas and key technology areas,and has a higher Of distinction.Secondly,calculate the correlation coefficients of PPR value and citation times,the number of claims,the number of patents of the same family and the PR value,and obtain the PPR algorithm and the PR algorithm,the citation times ranking is consistent,and the PPR algorithm is more recognizable than the PR algorithm Highly cited patents can identify key patents more than the number of times cited.In addition,the results of the PPR algorithm are more reasonable from a subjective perspective.Finally,the LFM algorithm,the PLFM algorithm,and the LLCDA algorithm are compared according to the expansion module degree.Detailed analysis proves that the PLFM algorithm improves the problems of excessive overlap and infinite loop in the LFM algorithm.