Research On Sentiment Classification Of Chinese Short-texts Based On Deep Learning

With the advent of Web 2.0 and 5G era,natural language processing has become the key of current big data research,especially sentiment classification of network text data,which is widely used in network recommendation,commodity marketing and various elections.In this paper,we mainly investigates emotional classification for the text data.First,we analyze the sentiment tendency of class balanced dataset for Chinese shorttext on the Internet by using convolutional neural network model and long-short memory network model.Compared with the support vector machine model and naive Bayesian model for traditional machine learning algorithm,the optimal sentiment classification model is obtained.Second,a Bi-Classification Balanced Cross Entropy Loss Function Focal Loss-2constructed on the class unbalanced dataset,which is adapted from Focal Loss Function.Experiments show that LSTM-word2 vec model has a highest classification accuracy reaching 93.13%;CNN-word2 vec model has the shorter training time;Focal Loss-2function rises 4% compared with the commonly used cross-entropy model when the scale of positive samples is not of large.Statistical tests show that the Focal Loss-2 function is better for classification performance on unbalanced data sets than the previous model.