Patent Mining Method And Application Based On WEW-LDA Model

At present,the society has entered an unprecedented period of rapid development,and scientific and technological innovation activities have shown a blowout trend of growth,becoming the first driving force to promote national economic growth and enhance national competitiveness.At the same time,with the continuous improvement of technology iteration speed,the competition in various industries has become increasingly fierce,and innovation has become the decisive factor of enterprises for the development and the maintenance of market competitiveness.Patent literature contains 90%?95% of the latest scientific research achievements worldwide every year,so mining information related to technology development from massive patent data to sort out the development status of technology system and reasonably predict the future trend can provide help for enterprises and countries to formulate innovative development strategies.Traditional patent research mostly focuses on measurement statistics,which ignores the important information hidden in the text.In the research of patent text mining that has emerged in recent years,there are problems such as the large dimension of feature extraction of patent text and the insufficient consideration of semantic relevance in the text,so it is not possible to complete the in-depth and systematic analysis of patent literature.Based on the quantitative analysis of the external structured information of patents,this paper constructs a patent-mining method for the intermediate level of the topic and the micro level of the abnormal patent for the unstructured data of the patent.First,the LDA model of unsupervised machine learning is introduced,and then combines the unique features of the patent text to improve the traditional model.After that the WEW-LDA model is proposed,which can achieve automatic recognition of technical topics in massive patent texts,and experiments have proved the advantages of this model in improving the readability of topics and the rationality of division.Then,based on the topic data output by the WEW-LDA model,a visual knowledge map such as the theme network and patent map is constructed.Among them,the topic network can analyze the external correlation between various technical topics based on the global text semantic information,and through the topic network,the structure of the technical system in the field can be clearly displayed.After the network is expanded in the time dimension of the life cycle,combined with social network analysis and link prediction algorithms,it is possible to mine the high-impact key topic transfer process,the overall evolution of technology and the possibility of association between future technology topics.As for the patent map,it can be used to detect abnormal patents containing new technologies.By screening,sorting and summarizing abnormal patents,it is possible to make a reasonable prediction of the future development of technology.Finally,the method system proposed in this paper is applied to the field of AGV.And a systematic analysis of the core topic transfer process,technology evolution,technology combination opportunities and future development trends in this field is carried out,which can provide guidance for the innovation development planning of AGV-related enterprises and national governments.