An improved wrapper-based feature selection for efficient opinion mining

Abstract

Opinion mining analyses people’s opinions, evaluations, sentiments, attitudes, appraisals and emotions to entities like products, organizations, services, issues, individuals, topics, events and their attributes. It is a large problem space having high feature dimensionality. Feature extraction is important in opinion mining as customers do not usually express product opinions totally, but separately based on individual features. Two tasks should be accomplished in feature-based opinion mining. First, product features on which reviewers expressed opinions must be identified and extracted. Second, opinion orientation or polarities must be determined. Finally, opinion mining summarizes extracted features and opinions. In this work a novel wrapper based feature selection mechanism using concept based feature expansion is proposed. The wrapper based technique uses the principles of evolutionary algorithms.