Research On Intelligent Retrieval Of Patent Infringement Based On Natural Language Processing

Patent is the most effective carrier of technical information and include more than90% of the world's latest technological achievements,which plays a vital role in the protection of intellectual property.With the continuous increase of the number of patents and patent infringement litigation,the research of patent infringement retrieval has become one of the hot topics in the field of information science.The traditional patent infringement retrieval mainly search patent from the patent retrieval system through the construction of retrieval,and then manually select the patents which have the risk of infringement.It's not only time-consuming but also easily affected by subjective factors.Therefore,it has important practical significance to study the intelligent retrieval algorithm of patent infringement.This paper introduces the concept of patent infringement,text preprocessing and similarity calculation in the patent infringement retrieval,and focus on the detection algorithm of patent infringement which is the core of patent infringement retrieval system.This paper puts forward the corresponding solutions to the problems in current research of patent infringement,such as the unreasonable choice of features and the insufficient use of the information of the patent claim.The main work of this paper is as follows:(1)As the weak expression ability of key words in the process of Chinese patent infringement retrieval,a algorithm of patent infringement detection based on the calculation of three tuple feature coverage is proposed.In this algorithm,the patent claim is extracted as a set of three tuples,and use the word vector and HowNet to calculate the semantic similarity between the three tuple features.Through the improvement of the patent technology feature set covering algorithm,we can effectively improve the ability to identify the patent infringement.The experimental results show that the proposed method achieves better detection performance and accuracy.(2)Because of the bad stability of the dependency parser and the low precisionof retrieval in method patent.a new algorithm based on sentence similarity calculation is proposed.This algorithm take the sentence as the smallest unit of computation,build the tree structure according to the characteristics of semi-structured for the patent claim,and design a tree matching algorithm to calculate the degree of patent infringement based on the rule of patent infringement judgment.Compared with the existing algorithms,the algorithm has some advantages.(3)In the Java platform,this paper adopts object-oriented thinking to design and implement the total modules of patent infringement retrieval system,such as intelligent database updates,preprocessing,preliminary retrieval,infringement detection and so on.Two kinds of detection methods are implemented in infringement detection module and other modules also improve the traditional methods.