Sentiment Analysis Based On Semi-Supervised Learing For Reviews On Social Media

With the rapid development of the Internet,people prefer to express their views on hot issues on social media.Sentiment analysis for reviews on social media has gradually become a hot research topic.Traditional sentiment analysis methods are mostly based on supervised learning,which requires sufficient initial sentimentlabeled data.However,labeling data requires a lot of manpower and time,which is not feasible in practice.To handle the scenario of insufficient sentiment labeled data,a semi-supervised sentiment analysis method based on dynamic threshold and multi-classifier is proposed.The unlabeled data are auto-labeled and filtered in an iterative way based on the proposed dynamic threshold algorithmin in this method,expanding the labeled training data.Speciallly,the threshold decreases with the increase of iteration times,where the higher threshold in the early iteration can ensure the quality of data and reduce the accumulation of errors in iteration,and the lower threshold in the later iteration can ensure the labeled data in the final training set is enough.In addition,the method also improves the traditional multi-classifier model and a weighted voting strategy is proposed.In particular,the product of the classifier's reliability on the prediction category and the predicted sample's reliability is taken as the final voting weight,which improves the generalization ability of the model.The experimental results show that the proposed sentiment analysis method has better effect on sentiment analysis comparing to other methods,proving the validity of the dynamic threshold algorithehm and weighted voting strategy strongly.Moreover,the sentiment prediction effect of deep learning model based on Long Short-Term Memory is better than that of machine learning model based on Support Vector Machine.It further illustrates that the sentiment analysis method based on deep learning can mine more abstract emotional features,obtaining better sentiment prediction effect.