Convolution Neural Network Based Patent Infringement Detection Method

Abstract

With the development of intellectual property rights in recent years, the number of patent applications has been increasing.At the same time, the number of patent infringement cases has also increased.When there is infringement between patents, the traditional method is for patent examiners to manually search for infringing features to determine whether there is infringement between patents according to the patent law.Since a patent is a complex semi-structured text and involves a wide range of fields, most of the current infringement detection methods cannot determine the infringement features well, and most of the methods only study one-to-one patent infringement and do not solve the problem of one-to-many patent infringement well.In order to solve the above problems, a patent infringement detection method based on convolutional neural network is proposed.The method extracts and represents infringement features from patents, patent claims and independent patent claims respectively, represents patents by different patent text vectorization methods, combines and filters features based on convolutional neural networks so as to obtain semantic information of different abstraction layers of patents, and finally tests the evaluation model on a one-to-many patent infringement data set.The results show that the model has greatly improved the infringement detection accuracy.