Research On Text Sentiment Classification Method Based On Deep Learning

Sentiment classification is one of the hot topics in information retrieval and data mining.It is aimed at finding out the attitudes and emotional tendencies expressed in subjective documents,and has great value on scientific research and application.In recent years,Sentiment classification has got wide attention and rapid development.So far,researchers have proposed many effective text sentiment classification algorithms.Among them,deep learning method has become an important method to solve text sentiment classification problem because of its excellent feature extraction ability.However,the deep learning models in the existing research ignores the effective utilization of the existing emotional resources and characteristics,and it can't take full advantage of the sequence characteristics of text data to improve the classification performance.To address this issue,this thesis focuses on the following tow research works:(1)To solve the drawback that it is insufficient for the deep learning method to extract features in the text sentiment classification task,this paper proposes a text sentiment classification algorithm based on the double channel convolutional neural network with extended features and a dynamic pooling.Firstly,the new algorithm summarized an extended text feature by combining with various word features that can influence the sentiment orientation of text,such as emotional word,part of speech,adverb of degree.negative word,punctuation,etc.Secondly,the word vector feature and the extended text feature are used as two individual channels of the convolutional neural network,and a new dynamic k-max pooling strategy is adopted to improve the efficiency of extracting feature.The results of performing text sentiment classification on numbers of standard English datasets demonstrate that the proposed algorithm has better classification efficiency than traditional convolutional neural network algorithm with single channel,and is highly competitive compared with some elitist text sentiment classification algorithms.(2)In order to improve classification performance by taking advantage of the sequence characteristics in text data.this paper proposes a text sentiment classification algorithm based on Long short-term memory neural network(LSTM)and attention mechanism(AM).Firstly,each text data is divided into several clauses by punctuations,which are inputted into the text sentiment classification model based on double channel convolutional neural network(TSCD-CNN)to extract the local features of the each clause.Then,the extracted features are input into LSTM in order to mine the sequence characteristics in text data,and an attention mechanism is employed to embody the emotional contribution differences of different sentences.Finally,the final classification model is obtained by training on training datasets.The experimental results on three standard English datasets demonstrate that the new algorithm can get better classification efficiency and classification performance compared with other six algorithms.