PSO-asent: Feature selection using particle swarm optimization for aspect based sentiment analysis

Abstract

The amount of user generated online contents has increased dramatically in the recent past. The phenomenal growth of e-commerce has led to a significantly large number of reviews for a product or service. This provides useful information to the users to take a fully informed decision on whether to acquire the service and/or product or not. In this paper we present a method for automatic feature selection for aspect term extraction and sentiment classification. The proposed approach is based on the principle of Particle Swarm Optimization (PSO) and performs feature selection within the learning framework of Conditional Random Field (CRF). Experiments on the benchmark set up of SemEval- 2014 Aspect based Sentiment Analysis Shared Task show the F-measure values of 81.91% and 72.42% for aspect term extraction in the laptop and restaurant domains, respectively. The method yields the classification accuracies of 78.48% for the restaurant and 71.25% for the laptop domain. Comparisons with the baselines and other existing systems show that our proposed approach attains the promising accuracies with much reduced feature sets in all the settings.