TEXT BASED SENTIMENT ANALYSIS OF PRODUCT REVIEWS

Abstract

With advent of web 2.0 people started sharing their opinions in social network. The social media helps in communicating
with public and provides a clear platform to share the views about the product. This has led to different ways of analyzing
the user reviews. Sentiment analysis is one of the wide spread area which helps in identifying the same. We proposed a
sentiment classifier which recognizes the opinion word based on linguistic analysis. This analysis is done before the
preprocessing stage so as to filter outliers and extract only the necessary words. In against the existing approach, the time
consumed in this method is considerably reduced as the reviews are analyzed in the initial stage. This method analyzes the
sentiment of the reviews posted by customers in online portals by taking the bigrams into account. The relationship between
bigrams is identified to know the wavelength of the user’s intention. The prioritized bigrams are chosen for every review
such that it qualifies the root word and the root word itself. To inculcate we have implemented a different theoretical model.
The data set we have taken for our experiment is a collection of 25,000 reviews from Cornell. The model was experimented
with different training sets where the accuracy and precision measures shows a marginal increase. The results of our
approach can be used for predicting the results in future as per the market specifications and future models. The accuracy,
precision and recall were the metrics that were used to identify the quality of our methodology. Thus the study on bigrams in
reviews yields an added value in sentiment analysis.