Component analysis of a Sentiment Analysis framework on different corpora

Abstract

Sentiment Analysis (SA) is the computational study of people's opinions about certain topics. With the massive growth of web 2.0 technologies, many sources of data and corpora are available for SA. There are some recent frameworks proposed in this field that can deal with different corpora. This paper presents a component analysis of recently proposed sentiment analysis framework. The framework components are divided to three stages, each of which contains many alternatives. The first stage is the text processing which include “handling negations, removing stopwords, and using selective words of part-of-speech tags”. The second stage is the feature extractions which are “unigrams and bigrams”. The third stage is the text classification which was done using “Naïve Bayes and Decision Tree” classifiers. It is important to analyze the components of the framework to configure which scenario is better for each corpus used. The analysis is enhanced by applying the framework components on the benchmark corpus movie reviews in addition to the prepared corpora from online social network sites and a review site. The results show that applying all the stages of text processing techniques ultimately decrease the classifiers' training time with no significant penalty in accuracy. The results also show that “Naïve Bayes” gives higher accuracy in case of balanced benchmark corpus while “Decision tree” classifier is better for imbalance data from social network.