Aspect based sentiment analysis for review rating prediction

Abstract

Aspect based sentiment analysis consists of aspect and sentiment extraction, and determination of the sentiment's orientation. In this paper, we propose a system to extract the aspect sentiment pair and compute the rating for each grouped aspect. Our approach starts with selecting the subjective sentences in the reviews. Then, it extracts aspects and opinions from the sentences, and determine the orientation of the sentiment. Aspects are clustered together by using WordNet with the prior knowledge of the aspect categories. Finally, it measures the rating for each aspect category. We evaluate our approach by three criteria: precision, recall, and F1-Measure. The subjectivity classification model and extraction model has F1-Measure value of 0.783 and 0.794 respectively.