Text Sentiment Analysis Based On BiGRU-Attention Model

With the rise of major E-commerce platforms,people tend to buy goods and evaluate commodities online,and become a trend.The amount of commenta ry data is growing by hundreds of millions every day.When the data reaches a certain level and these seemingly unruly or less relevant comments are subdivided into a certain industry or a commodity,people's general views of a certain industry or a commodity will can be shown by the analysis of these data.People's general views of a certain industry or a commodity is of great value in analyzing commodities and developing commodities and predicting sales prospects.So text sentiment analysis has become popular in recent years.The original text sentiment classification algorithm is based on machine learning algorithm.But with the increase of data volume,machine learning algorithm can not adapt to the demand of fast processing data.The neural network model is suitable for large data analysis and has begun to be widely used in the field of natural language processing(NLP).CNN was first used to classify text emotions.Because CNN could not learn the characteristics of context information.RNN began to be used in text sentiment classifications.The problem of RNN gradient diffusio n made LSTM,GRU,Bi LSTM,Bi GRU and other variants become popular.Later,Attention mechanism was proved to be effective in highlighting local information.Researchers began to try to mix the neural network model with the attention mechanism.And the experimental results proved the effectiveness of the mixing models.Aiming at the problems that the current model does not consider the possible spelling errors of comments,the length o f word vectors and the long training time of Bi LSTM which is widely used,a nd the inadequate extraction of text information,this paper proposes a text sentiment classification model based on Bi GRUAttention,which imports textblob package to correct the p ossible spelling errors before preprocessing and fills in the input layer with the pad_sequences.Bi GRU is mainly used to learn the text context information and Attention mechanism is used to highlight the key word vector information.The Bi GRU-Attention model adopts six-tier structure.It extracts the features of text word vectors through the input layer and the neural network layer,puts the key information of w ord vectors into Attention mechanism,puts it into Dropout layer to prevent over-fitting,passes through full connection layer,and finally puts it into the Softmax layer to classify text emotions.The Bi GRU-Attention model is tested on four data sets,which verifies the validity of the Bi GRU-Attention model.The paper has 33 figures,14 tables and 52 references.