Research And Implementation Of Patent Recommendation Method Based On Text Similarity

Scientific and technological innovation can improve productivity and promote social progress,and intellectual property rights,as the "umbrella" of scientific and technological innovation,can encourage scientific and technological workers to innovate.Patent is an important carrier of knowledge and innovation achievements,and also reflects the competitiveness of enterprises in the market.How to make patent recommendation efficiently has become a hot topic of research and an important business requirement for users and enterprises.In recent years,natural language processing techniques have been widely used in various fields.In this thesis,we use such techniques to analyze patent and enterprise characteristics and recommend suitable patents to enterprises,so as to promote the conversion efficiency of patent property rights and improve productivity.This thesis collects and analyzes enterprise and related patent datasets,proposes the keyword-based Inter-TF-IDF enterprise patent recommendation method and the semantic-based Contrast BERT patent recommendation method,and implements a patent recommendation system based on the above two methods.The main contents of this thesis are as follows.1.Using crawlers to crawl the national laser enterprises and laser class patent data information,organize and establish the corresponding database.2.The method of enterprise patent recommendation based on Inter-TF-IDF algorithm.The traditional TF-IDF algorithm ignores the distribution of lexical items within a certain category of text and the distribution of lexical items among categories in text matching,thus ignoring the differentiated attention of enterprises to patent information,resulting in the recommendation method based on TF-IDF algorithm cannot meet the personalized needs of enterprises.To address such problems,this thesis proposes the Inter-TF-IDF algorithm based on the distribution characteristics of word items in categories.And a comparison experiment is conducted,and the results show that the recommendation method based on Inter-TF-IDF algorithm has a greater improvement in accuracy.3.Enterprise patent recommendation method based on Contrast BERT model.Since the patent text data contains a large number of similar words,the keyword-based patent recommendation method has certain limitations and cannot meet the needs of users,so the Contrast BERT model that distinguishes similar texts in the semantic space is proposed.Through comparison experiments,the results show that the recommendation effect of the recommendation method based on this model has a high accuracy rate.4.An enterprise patent recommendation system based on Inter-TF-IDF algorithm and Contrast BERT model.In order to meet the needs of different users,this thesis implements a keyword-based and semantic-based patent recommendation system.By using this system,users can obtain accurate and comprehensive patent recommendation results according to their own needs.