Sentiment analysis by POS and joint sentiment topic features using SVM and ANN

Sentiment analysis using the part-of-speech (POS) tags and the joint sentiment topic features is a novel idea. As the sentiment analysis requires effective selection of features which are utilized in the determination of sentiment. In this paper, the POS tagging is performed by using hidden Markov model where the unigrams, bigrams and bi-tagged features are extracted. Similarly, the nonparametric hierarchical Dirichlet process is employed to extract the joint sentiment topic features. The extracted features are combined together in a linear fashion in order to effectively select the best feature subset. The best features are selected based on maximum relevance and minimum redundancy mutual information of the feature subset. The mutual information is used to measure the relevance between features and sentiment analysis decision. The maximum relevance and minimum redundancy mutual information remove the redundant features by considering the mutual information between features. Feature selection is carried out by these fitness conditions using firefly optimization algorithm. Then, the chosen feature subset is employed in the classification process which is performed using support vector machine and artificial neural networks. Thus the proposed sentiment analysis method provides more accurate sentiment recognition. Experimental results show that the proposed sentiment analysis method improves the accuracy and reduces the training speed for sentiment analysis.