Particle Swarm Optimization Based Feature Selection and Summarization of Customer Reviews

The steady growth of e-commerce has led to a significantly large number of reviews for a product or service. This gives useful information to the users to take an informed decision on whether to acquire a service and/or product or not. Opinion mining techniques are used to automatically process customer reviews for extracting feature and opinion in a concise summary form. Existing feature based summarization system uses dependency relations and ontological knowledge with probabilistic based model to generate the summary. To enhance the accuracy of summarization, fitness proportionate binary particle swarm optimization (FBPSO) based feature selection is proposed. The BPSO could efficiently search for subset of features using fitness sum based on the multi-objective function. The FBPSO overcome the problems of traditional BPSO as it focus much on the overall performance of a particle as a whole and it does not pay attention on every single feature. The multi-objective function used in FBPSO is based on dominance, mutation and crowding factor to generate an efficient summary. The performance of the system is measured using the Recall-Oriented Understanding for Gisting Evaluation (ROUGE) toolkit. Experimental results show that the proposed approach of summary generation using multiobjective BPSO algorithm outperforms the traditional probabilistic model. Keywords— Feature selection, Multi-objective, Binary Particle Swarm Optimization, Summarization