Patent Document Semantic Retrieval Based On Character Convolutional Neural Network

With the advancement of economic globalization,science and technology have become more and more important in countries and enterprises.As a carrier of science and technology,patents have become an important measure of the country's innovation.Patents are an important part of intellectual property,and patented technology has become a core element of competition between countries or companies.Nowadays,the number of patent applications is increasing day by day.In the face of massive patent information,how to extract the scientific and technological information contained therein to provide strategic support for the development of the country and the enterprise is the focus of research at home and abroad.Different International Patent Classification Numbers(IPCs)have been assigned according to the differences in the technical content of patents in order to find the target patent information more efficiently.However,the current patent classification mainly stays in manual operation,and the classification efficiency is low,the cost is high,and the classification result does not have a uniform control standard.Therefore,studying the semantic retrieval of patent texts has important social realistic significance.The experimental data of this thesis was selected from the Baiten patent library,and the patent text data was segmented by Jieba tool.The convolutional neural network was used to convolute the patent text data,extract the characteristics of the data and classify it,and then use Word2 vec to transform the patent into word vector set,and the tested text is matched with the vector set of category words to retrieve similar patent texts.The main work is as follows:1.Analized the research significance of the semantic retrieval algorithm for patent texts,the current research status at home and abroad,and the shortcomings of the current traditional semantic search algorithms for patent texts.2.Studied Several commonly used text semantic analysis algorithms,including the basic theory of convolutional neural networks,the transformation of word vectors,and the working principle of word2 vec tools.3.Verified the validity of semantic retrieval of patent text based on character-level convolutional neural network model,and analyzed the selection methods of each parameter.Proposed a combination of character-level convolutional neural network model and word2 vec to retrieve the semantics of patent texts.4.Designed and implemented the patent text semantic retrieval based on character level convolutional neural network.Experimental results show that the proposed algorithm can effectively improve the recognition accuracy compared with the existing patent text semantic recognition algorithm.