An Unsupervised Hierarchical Rule Based Model for Aspect Term Extraction Augmented with Pruning Strategies

Abstract

Abstract Sentiment Analysis is an interesting field in text analytics which has emerged from an individual’s curiosity to know what others think or feel about an entity or experience and hence take a decision. Aspect level sentiment analysis is a sub-field which operates at a more fine grained level to meet the demand of the end users who are no longer satisfied with the overall sentiment about the entity but are keen to know what features of the entity are matters of concern and what inturn is the sentiment reflected on each of these features/aspects. Aspect term extraction is the first and the most difficult step in aspect level sentiment analysis which attempts to discover the features of the product that are discussed about in the reviews. The proposed approach presents an unsupervised hierarchical rule based approach for aspect term extraction oriented towards a high recall and augmented by pruning strategies for filtering the false positives. The proposed model has reported an appreciable recall of 81.9 and 68.7 on Restaurant and Laptop domains respectively on SemEval 2014 dataset and has also been compared with the state of art models.