Research And Application Of Deep Learning Based Patent Recommendation Algorithm

According to the statistical report of the five largest intellectual property offices in the world in 2018,by the end of 2017,there were 13.6 million valid patents globally,and China had more than 2 million valid patents.How to efficiently mine patents that users may be interested in from these patent data to assist researchers in researching and writing patents has become a huge challenge in the field of patent research.Recommendation algorithms are the main branch of data mining and are widely used in different fields.Currently in the field of patents,traditional recommendation algorithms have the problem of single-sided feature portrait,and the problem of low accuracy of similarity matrix calculation between users in a sparse interaction matrix.Deep learning models are widely used in the field of natural language processing,and them can more accurately represent the characteristics of patent documents.Therefore,this thesis uses deep learning models to improve patent recommendation algorithms and applies them to patent writing assistance software,so as to improve the quality of patent writing and the value of patent.This thesis mainly focuses on three aspects: user interests based patent recommendation algorithms,collaborative relationships based patent recommendation algorithms,and patent writing assistance systems based on recommendation algorithms.The main work of this thesis is as follows:(1)In terms of users' interests,aiming at the problem that the word frequency vector of feature portrait is single-sided,user portrait based patent recommendation algorithm is proposed.This algorithm obtains Word2 vec deep learning model by training patent data source corpus,then can represent feature portrait more accurately and comprehensively by the deep learning model,finally,compared with the traditional content recommendation algorithm in the field of patent,this algorithm has advantages.(2)In terms of collaborative relationships,aiming at the problem of low accuracy of similarity matrix calculation between users in a sparse interaction matrix,deep semantic similarity based patent recommendation algorithm is proposed.This algorithm obtains Doc2 vec deep learning model by training patent data source corpus,then constructs the semantic similarity matrix between patents through the deep learning model,and then combines completion strategies to complete the interaction matrix,so as to more accurately calculate the nearest neighbors of the target user,finally,compared with the traditional user-based collaborative filtering recommendation algorithm in the field of patent,this algorithm has advantages.(3)Aiming at the problem that patent researchers cannot use massive patent data to research and write patents more efficiently,this thesis designs and implements a patent writing assistance system based on the recommendation algorithm.The system includes recommendation engine modules,patent writing assistance modules,system modules,and user interest collection modules.This thesis implements a parallel hybrid recommendation engine that combines user interests and user collaboration,so as to recommend patents comprehensively and accurately and realize intelligent retrieval,then applies it to patent writing assistance software to assist patent researchers to research and write patents more efficiently,so as to improve the quality of patent writing and the value of patent.