Convolution Kernel Adaptive Text Classification Algorithm Based On Multi-channel Feature Representation

With the rapid development of science and technology,more and more text-based information appears in forums,blogs,post bars and other media.On the one hand,the rapid spread of text information enriches people's life.On the other hand,it also brings some difficulties to the supervision of information.Because a large amount of negative and false information is full of Internet and other media,how to classify text information effectively and accurately has become an important research topic in the field of computer science.In recent years,thanks to the rapid development of natural language processing,text classification technology has also made a breakthrough progress.Various text classification models based on deep learning technology emerge in endlessly.However,these models still have shortcomings: 1)the traditional text classification models use word vectors from a single source as the input of the model,and the feature representation of the text sequence often forms a single channel through various fusion ways,which makes the text semantic features not rich enough;2)although the length of sequence feature fragments that play a decisive role in text classification is different,the existing text classification models still have the problem that between the window size of convolution kernel and the length of key fragments in the sample sequence have low matching degree,which can not accurately capture the influential key text features,resulting in a large amount of redundancy of text information,making irrelevant information affect the classification accuracy and the performance of text classification model cannot be improved.Based on the above problems,this paper proposes a convolution kernel adaptive text classification algorithm model based on multi-channel feature representation.In order to solve the problem that a single channel formed by word vectors from a single source is not rich in text feature representation.Firstly,this paper proposes a multi-channel feature representation method.In this method,word vectors from different sources are used as the input of two bidirectional long short-term memory networks,and the output of the two networks at each time is stacked vertically to form a multi-channel text feature representation,so as to capture the information of the context at the same time and enrich the semantic information of the feature representation.Secondly,in order to solve the problem of low matching degree between convolution kernel window size and target sequence length in traditional text classification models,a convolution kernel adaptive text classification method is proposed by integrating the attention mechanism into the multi-channel feature representation.In this method,the multichannel feature representation of convolution check with different sizes is used to extract features,and then the convolution features with different particle sizes are weighted in the way of attention mechanism,so that the convolution features that have decisive effect on classification have larger weight,so as to achieve the purpose of self-adaptive convolution kernel width.Finally,this paper compares and analyzes proposed text classification model on six datasets.The text classification model proposed in this paper has certain generality.Although the English data set is used in the experiment,it can be easily migrated to other text classification scenarios,so it has certain theoretical research and value.At the same time,the experiment shows that the classification model proposed in this paper can effectively classify the news datasets,movie reviews datasets and other area datasets,so it has a certain practical significance for practical application.