Research On New Patent Retrieval Method Under Deep Learning

Intellectual property is a concept of the rule of law.It refers to the exclusive right that people enjoy in accordance with the law in the results of their intellectual labor,which the state grants the creator to enjoy intellectual achievements for a certain period of time[1].Intellectual property with patents as an important component has become the hot topic of competition between the state and the state,enterprises and enterprises.Patents carry the most core technical points and worthy of research and preemption,both for the country and the enterprise.Referring to the data released by the authority of the State Intellectual Property Office of China,as of March 2017,the number of patent applications for inventions in the world's top five bureaus has reached 2.6 million,and the number of patents granted has also reached 1.1 million[2],which indicates that people pay more attention to the protection of their own intellectual property rights.Faced with such a huge amount of patent data,how to effectively use it has become a problem that related companies and scholars have been trying to solve.For patent-licensing organizations,it is necessary to find whether the current applied patent is the same or similar with other patents in the vast amount of patent data texts,which is related to whether the current applied patent can be authorized.For enterprises,it is necessary to find patent collections in related fields to avoid technology infringement.For patent-writing units or individuals,it is necessary to avoid overlapping between the patents that are written and existing patents,in order to increase the possibility of patent authorization.It can be seen that patent retrieval runs through all cases of patent application.There are already many researches on patent retrieval.Many important international organizations and conferences have formed corresponding workshops for the analysis and research of patent documents,such as SIGIR(Special Interest Group on Information Retrieval),ACL(The Association for Computational Linguistics),NTCIR(The Japanese National Institute of Informatics Testeds and Community for Information access Research project),and so on.Thus,many excellent algorithm models have been proposed.However,the query performance of patent retrieval is not good,and the recall rate and accuracy rate still need to be improved.Aiming at the current research status of patent retrieval,this paper proposes a novel patent retrieval methos based on deep learning,which uses the ability of deep learning to handle massive amounts of data and the ability to learn automatically.This paper aim to further increase the recall rate of patent retrieval and ensure the precision at the same time.Through the deep learning model,the patent text set is trained into the expression form of word vector,giving each keyword a unique vector,and the keyword correlation calculation is converted into a calculation between vectors.Then keywords are mapped to a node in the graph.The edges between the nodes and the nodes are represented by the values calculated between the vectors.A dense subgraph algorithm is proposed to obtain an extended word set,and then using these keywords to search in the database.Finally,this paper will sort the set of search results using the patent document ranking model.