Part of speech features for sentiment classification based on Latent Dirichlet Allocation

The input data used in the sentiment analysis process by using machine learning generally is Bag of Word (BoW). However, the input data using BoW is not enough to improve the machine learning in defining the polarity in a document. Therefore, need input in the form of more specific feature so that it is capable to give the more maximal result. Part of Speech (POS) is one of the techniques to create the more specific feature in a document. By using the POS-based feature in a document, then the occurrence of the word class like adjective or negation can be detected. The adjective and negation are the main sign of the sentiment or opinion in a document. This study is aimed to use POS technique to conduct feature selection. The result of the POS-based feature process will be the input for sentiment analysis process by using Latent Dirichlet Allocation (LDA) method. The result of this research showed that the document which has passed the POS-based feature process can give accuracy score higher with the difference about 7.8% than the document without feature selection process of POS.