Research And Implementation Of Text Sentiment Classification Algorithm Based On Deep Learning

With the development of science and technology,especially the development of mobile Internet,the Internet has produced many user opinions and feelings of text data,and every year in the rapid growth,if using text sentiment classification technology mining the emotion information of these data,it will be useful to understand people's opinion of the public events,and it will be useful for the company to understand the product and user.Therefore,the study of text sentiment classification has important social significance and commercial value.At present,the commonly used emotion classification methods are rule-based and machine learning based.But the rule-based method are dependent on the quality of the dictionary and rule,and the machine learning based method are dependent on manual designed feature representation.Rule-based and machine learning algorithms often can't classify emotional text from the semantic level,thus cannot deal with complex relations.In recent years,deep learning technology has been widely used in the field of Natural Language Processing.Deep learning is called representation learning,which can automatically learn the feature representation of the text,and can deal with context dependency.Therefore,this paper will study how to use deep learning technology to further improve the accuracy of text sentiment classification algorithm.As recurrent neural network can handle long distance dependency and capture the semantic information of input text,and convolutional neural network is often used to extract the n-gram features of input text.In this paper,a novel text sentiment classification model(BL_CNN)combined with recurrent neural network and convolutional neural network is proposed,which consists of word embedding layer,bidirectional recurrent neural network,convolutional neural network and output layer.BL_CNN first gets word vector representation of input text by embedding layer,then gets input text context representation by using bidirectional recurrent neural network,then gets the representation of the input by convolutional neural network,finally gets emotion categories by output layer.According to recurrent neural network and convolutional neural network are prone to overfitting,while the standard dropout is not effective and stable to prevent recurrent neural network from overfitting,BL_CNN introduced a more suitable dropout for recurrent neural network called variation dropout.Experiments show that the BL_CNN model can achieve the best results in multiple data sets.Compared to the accuracy of the currently known best model,the BL_CNN model has a 1.3% improvement in the Stanford Sentiment Treebank data sets,a 0.6% increase in the MR data set,a 0.5% increase in the Chinese hotel review document data sets,The accuracy in the IMDB data set is second only to the best model.The experimental results of the above data set show that the combination of recurrent neural network and convolution neural network can further improve the accuracy of text emotion classification algorithm.