Research On Short Text Sentiment Analysis Of GRU Based On Attention Mechanism And SVM

In recent years,with the development of the Internet industry,a large number of online texts have been generated.These texts are abundant in information,among which the emotional information has played a significant role in analyzing public opinions,designing marketing strategies,predicting share prices,etc.In this sense,an investigation into the sentiment in online texts is of great value.With the increase of textual data,more manual efforts are needed for labeling and categorizing.Meanwhile,the sparsity of short texts also makes the traditional methods undesirable.Aimed at dealing with the above problems,the current study proposed a GRU(Gated Recurrent Unit networks)short text analysis model based on SVM(Support Vector Machine)and attention mechanism.The innovations are as follows:(1)To solve the problems emerging in emotional dictionary and traditional machine learning,the current study proposed an analysis model in sentiment analysis of short texts on the basis of deep learning.Several typical structures of deep learning methods were analyzed and compared and a bidirectional GRU model based on attention mechanism was proposed for better context information and higher speed of model training.Making use of a word2vec model,the model was able to transform texts into vectors with meanings.Then the irrelevant features were weakened by stacking two unidirectional GRU models before and after,adding the multi-attention mechanism and inserting attention behind the input layer and in front of the classifier.The experiment proved that the improved model achieved better classification effect and the accuracy rate improved by 1.7%in the dataset of Chinese hotel reviews.(2)To deal with the feature sparsity of short texts,an Analysis model of sentiment in texts combined with SVM was proposed.During the experiment,the Bi-GRU model served as the feature extractor,SVM as the classifier and output of the Bi-GRU model made the input of the SVM model.The improved model made full use of the advantages of SVM(strong robustness and high accuracy)and achieved optimal performance.It was proved that the addition of attention mechanism and the change of classifiers improved the performance of the original GRU neural network model,and the classification accuracy rate reached 88%.The model had a higher accuracy compared with the traditional classification model.