A Supervised Approach to Aspect Term Extraction Using Minimal Robust Features for Sentiment Analysis

The instinct to know what others feel lays the foundation for the field of sentiment analysis which extracts opinion from text data and categorizes them as positive, negative or neutral. Beyond a report of the consolidated sentiment, the end-user is more interested to know what the product features that are talked about and what is the sentiment of the opinion holder towards each feature/aspect which leads to the task of aspect-level sentiment analysis. In this paper, the focus has been on the aspect extraction task of aspect-level sentiment analysis which extracts the features of the product that has been talked about in the reviews. The experiments have been reported on Bing Liu Customer Review Datasets consisting of five different categories DVD, Canon, MP3, Nikon and Cell phones. The strength of the model lies in the fact that a simple classifier that incorporates handling of imbalanced data, using a minimal set of robust features has been able to achieve comparable results with the state of art in aspect extraction task. The random forest classifier reported the best results across all domains with an F-measure ranging from 85.3 to 89.1.