基于深度学习的中文专利自动分类方法研究

Research on the Method of Chinese Patent Automatic Classification Based on Deep Learning

[目的/意义]面向当前国内专利审查和专利情报分析工作中对于海量专利分类的客观需求,设计了7种基于深度学习的专利自动分类方法,对比各种方法的分类效果,从而助力专利分类效率和效果的提升。[方法/过程]针对传统机器学习方法存在的缺陷,基于Word2Vec、CNN、RNN、Attention机制等深度学习技术,考虑专利文本语序特征、上下文特征以及分类关键特征,设计Word2Vec+TextCNN、Word2Vec+GRU、Word2Vec+BiGRU、Word2Vec+BiGRU+TextCNN等7种深度学习模型,以中国专利为例,选取IPC主分类号的"部"作为分类依据,对比这7种模型与3种传统分类模型在中文专利分类任务中的效果。[结果/结论]实证研究效果显示,采用考虑语序特征、上下文特征及强化关键特征的深度学习方法进行中文专利分类具有更优的分类效果。

[Purpose/significance]In order to meet the needs of classifying massive patent automatically in current patent examination and patent information analysis work,this paper studies a series of patent automatic classification methods based on deep learning and compares the classification effects.This will promote the efficiency and effectiveness of patent classification.[Method/process]Aiming at the shortcoming of traditional machine learning methods,7 deep learning models was designed,including Word2Vec+TextCNN,Word2Vec+GRU,Word2Vec+BiGRU,Word2Vec+BiGRU+TextCNN and so on.These models based on the deep learning technology,such as Word2Vec,CNN,RNN,Attention mechanism and so on and considered the characteristics of patent text word order,context features and other key features in classification.Selecting the‘Section’of main International Patent Classification(IPC)was as the class labels,the study classified the Chinese patents by above 7 deep learning models and 3 traditional machine learning methods.And there was a comparison about the effect of classification in different models.[Result/conclusion]The empirical research indicated that it reached the better effect of Chinese patent classification by using deep learning methods which considered the characteristics of patent text word order,context features and other key features in classification.