Text Sentiment Analysis Based On Improved SO-PMI Algorithm And Double Channel Convolutional Neural Network

With the explosive growth of text data on the Internet,text sentiment analysis becomes more and more critical.We can use text sentiment analysis to conduct public opinion analysis,trend prediction and other jobs.In particular,the mining of product review information is helpful for merchants to adjust their sales strategies in a timely manner,and buyers can buy more affordable products.For the task of text sentiment analysis,this experiment completed the following work:In the construction of sentiment dictionary,this paper studies the automatic construction of sentiment dictionary in the e-commerce comment text data set.First,TF-IDF algorithm is used to filter low frequency words and non emotional words.Then,we can use the ratio of the probability of positive tendency in all texts containing words to the probability of positive tendency in the whole text data set as the PMI value of positive emotion of the word.Similarly,we can get the PMI value of negative emotion of the word.We get the result of two values of subtraction and then multiply the square of word frequency to get the SO-PMI value of the word.According to the SO-PMI value,we can complete the automatic construction of sentiment dictionary.And the sentiment information in the sentiment dictionary is added to the word vector to get the sentimental word vector.the Convolutional Neural Network model is used to verify the sentimental word vector.The accuracy of the model trained by the sentimental word vector is 90.13%,and the accuracy of the model trained by the common word vector is 88.88%.The advantages of the sentimental word vector in the text sentiment analysis task are verified.In the construction of deep learning model,the Convolutional Neural Network model is used to determine the parameters of the model.On this basis,a Double Channel Convolutional Neural Network model is constructed,GRU and Attention Mechanism are added to the model to improve the accuracy of the model.The experimental results show that the accuracy of the deep learning network model based on the Double Channel Convolutional Neural Network and bidirectional GRU and Attention Mechanism is 96.62%,6.49%higher than the Convolutional Neural Network model,and 4.08%higher than the bidirectional LSTM network model.