Aspect-specific polarity-aware summarization of online reviews

Abstract

With the popularity of various social media platforms, the number of online reviews towards different products and services grows dramatically. Discovering sentiments from online reviews becomes an important and challenging task in sentiment analysis. Current methods either extract aspects without separating aspects and sentiments, or extract aspects and sentiments without separating sentiments according to their polarities. In this paper, we propose two novel probabilistic generative models (APSM and ME-APSM) to extract aspects and aspect-specific polarity-aware sentiments from online reviews. We applied our models to two data sets with three different experiments. Experimental results show that APSM and ME-APSM models can extract aspects and polarity-aware sentiments well. For the sentiment classification task, our models outperform other generative models and come close to supervised classification methods. © 2013 Springer-Verlag Berlin Heidelberg.