Comparative Analysis Of Patent Literature Based On Deep Topic Model

With the rapid development of science and technology,intellectual property is an important amount of competitiveness for enterprises,individuals,research institutes and universities.Protecting intellectual property from infringement is a necessary measure to protect its own competitiveness.For patent applicants,comparing patent documents with existing patent technology and avoiding similar parts to existing patent technologies can effectively improve the grant rate of patent.This article first studies the current patent comparative analysis methods and related applications.The current patent comparative analysis method is mainly based on text mining and visualization technology.Patent analysis based on text mining is a method that uses analysis tools to obtain meaningful knowledge information from natural language patent texts.It obtains important information by identifying and detecting important patterns from a large amount of text data.However,for patent texts,text mining cannot accurately represent technical concepts,and there are also significant limitations in identifying synonyms.Visual patent comparison and analysis technology is to display patent information in a visual way,which is beneficial to manual analysis of patent results.For example,patent maps or patent networks can be used to understand patent technology trends in specific areas.Based on the current technology,there are two main researches in this paper: one is the classification of patent documents based on deep topic models;the other is the one-to-many patent comparative analysis.The existing patent classification mainly adopts the traditional text classification.The text mining is used to extract the features of the text,and then the text is classified based on the extracted features.This paper uses a deep topic model to extract the topic distribution of patent,and then classifies the patent based on the topic distribution.This article introduces word embedding and topic embedding based on the classic topic model and fuses word embedding and topic embedding into the topic model through the attention mechanism.Word embedding and topic embedding take into account the position of words,which can supplement the semantic information between words.This one-to-many patent comparative analysis framework compares similar parts of the target patent and similar patent sets,and generates a comparatively readable summary.The experiments prove that the patent classification based on the deep topic model in this paper has better results than the traditional classification model.The one-to-many patent comparative analysis framework can effectively extract the technology features of patents and generate a highly readable summary.This is beneficial for patent applicants and patent review experts to make clear judgments on patent applications.