Research On Text Sentiment Analysis Based On Neural Network And Ensemble Learning

How to effectively represent text semantic features and mine the potential sentiment polarity of text has always been the focus of research in the field of text sentiment analysis.The traditional method based on sentiment dictionary often needs to manually construct a large number of sentiment vocabulary,and the accuracy of sentiment classification depends on the completeness of the sentiment dictionary.However,there are few mature emotional dictionaries,and it is time-consuming and laborious to construct an emotional dictionary.Tagged training data is the basis for the use of machine learning for sentiment analysis.Insufficient training data or severe bias will cause the classifier to fail.Using deep learning for emotional analysis can automatically capture data features through multi-layer neural networks,which improves the accuracy of text analysis,but a single classification model may have unstable classification.Aiming at the above problems,this thesis proposes a sentiment analysis method that combines neural network and ensemble learning.Firstly,two different types of Chinese and English datasets are selected,the text data are preprocessed and word vectors are trained,and feature extraction is carried out with the Doc2 vec method.Then use the support vector machine(SVM),the long short-term memory network(LSTM),the convolutional neural network(CNN)and the joint model of the convolutional neural network and the long and short-term memory network(CNN-Bi LSTM)to train model separately.In order to solve the bottleneck that complex models do not improve the classification accuracy,the idea of Stacking ensemble learning is introduced,and a heterogeneous ensemble model based on Stacking is proposed,which integrates four basic classifiers.And compared with other benchmark models mentioned in the reference,it is verified that the heterogeneous ensemble model based on Stacking can effectively improve the classification effect.The experimental results show that the heterogeneous ensemble model based on Stacking can effectively improve the accuracy of text sentiment analysis.The stacking-based ensemble model proposed in this thesis has a classification accuracy of97.66% on the IMDB dataset,and a classification accuracy of 94.24% on the Weibo comment dataset,which is better than other traditional classification models,demonstrating the feasibility and effectiveness of the proposed model.