Patent Search Based On Deep Learning

In recent years,with the rapid development of science and technology such as the Internet of things,robots and artificial intelligence,the trend of economic globalization is becoming increasingly obvious.The possession,allocation,production and application of knowledge or intellectual resources(including monographs,patents,trademarks,scientific papers,etc.)have become an important support for China's economic development,and the importance of technical knowledge has become increasingly prominent.The proportion of knowledge-based industries in the domestic economy keeps increasing,and intellectual property rights have become the focus of competition between countries and enterprises.As a typical intellectual property right,patent is an important index to measure technological innovation,and also a means of legal protection from technology infringement.Patent retrieval has become an essential part of the research and development process.The retrieval and analysis of massive patent documents is a comprehensive field.Its progress depends on the research of natural language processing,machine learning and information retrieval.As a new field of machine learning,deep learning aims to make machine learning more close to its original goal artificial intelligence.How to use the deep learning method,combined with natural language processing technology to search and analyze patents,has practical significance in the aspects of infringement analysis suggestions,competitor suggestions,market positioning suggestions,etc.For patent retrieval,according to the query given by the searcher,the patent text set related to the query is returned and sorted in descending order for the user to use.Traditional patent retrieval is mostly carried out by means of mechanical semantic matching,which usually can't reach the retrieval concept of the searchers.However,artificial intelligence technology is more and more used in patent retrieval.By analyzing and extracting the key information of the patent to be examined,and extending the corresponding semantic,the retrieval performance can be improved to a certain extent,while a single retrieval technology has some advantages limitations.In view of the above problems,based on the in-depth study of text representation,this paper mainly studies the fusion strategy of the two retrieval methods and query expansion technology to improve the performance.First,this paper uses the skip gram model of word2 vec to express the word vector and calculate the similarity,constructs the patent keyword network,and finds the sub graph satisfying the conditions from it to expand the query;then combining the two retrieval models based on DAE and DSSM,they express and calculate the similarity of the query and the text respectively,and then sort the results by weighted sum.Experiments show that the proposed method can improve the recall and accuracy of patent retrieval.