OACD-SA: Online Aspect Category Detection for Sentiment Analysis using Unsupervised Learning

The web is a vast repository of consumer reviews, but it is impossible to go through all reviews to arrive at a fair evaluation when we need to buy any product or service. A recommender system that can quickly categorize reviews is highly needed. A subtask of this system is to categorize the general aspects and check sentiments in consumer reviews to provide a proper suggestion or recommendation to any user. The proposed Online Aspect Category Detection system for Sentiment Analysis (OACD-SA) uses the SemEval 2014 Task 4 as a baseline for ‘Aspect detection’ and then goes on to use online reviews given by users as text corpus for any product or service. Unsupervised learning method along with association rule mining is used to discover aspect categories from a corpus. It uses a novel spreading algorithm to rate every feature which can then be used for evaluation of the overall rating for that product. The unsupervised learning method returns a F1-score of 68% for aspect detection and an aspect sentiment classification accuracy of 66% on baseline task, thus providing robust aspect categorization with a quick response time for any online product. Therefore, the proposed system can be to improve their product or service quality.