Research On Intelligent Patent Infringement Retrieval Based On Neural Network

As a vital part of intellectual property, patent has received a widespread concernfrom all over the world. Patent application is largely promoted by steadily soundintellectual property protection mechanism, which on the other hand induces theoccurrence of a large amount of patent infringement case. According to the data ofChina’s Intellectual Property Office, lately, there exists an increasing trend in patentapplication. At the same time, patent infringement case and the corresponding amountof compensation are increasing, especially the lawsuits between large enterprises.Therefore, based on research status in patent infringement retrieval, we adopt theneural network theory, and construct the Chinese patent infringement retrieval modelusing patent data, in order to provide an accurate, comprehensive retrieval result forR&D personnel from the aspect of avoiding patent infringement, and in case ofrepetitive studies and accidental patent infringement. On the other hand, we provide aquick and accurate patent examination method for patent examiners, and guaranteethe timeliness of the applied patents.The main contents include the acquisition of Chinese patent information and thecorresponding data processing, the construction of patent infringement retrieval model,the implementation of the retrieval system and evaluation of experimental effects. Thedata come from China’s Intellectual Property Office, both of authorized inventionpatents and invalid invention patents. The patents data are retrieved and downloadedfrom PatentEX, and the patent claims are further extracted from the patentspecification documents. In the phase of text preprocessing, first, the stop words listand technology field words list are generated on the basis of corpus, which are utilizedto filtrate the patent data. Secondly, on the basis of Fudan University NaturalLanguage Processing System, a special word segmentation algorithm is proposed todeal with the word segmentation process, namely, TextRank Algorithm, which issuitable for Chinese patent claims. Finally, the co-occurrence of patent number andkey words are generated by extracting the keywords and calculating the weight ofeach key word. In the phase of construction of patent infringement retrieval model, weadopt SOM neural network algorithm, fulfill the clustering analysis of patent dataset in the platform of Matlab, and after a further step of classification, the highlysuspected infringed patents are determined. After evaluation of the results ofclustering process and the comparative analysis, it reveals that the appoach is feasibleto simplify the retrieval process and improve its accuracy. In addition, the system ofthe model above is achieved on the basis of corpus, SOM model and evaluationalgorithm, and the interactive interface are optimized from the aspects of the users.