FEATURE SELECTION USING MODIFIED ANT COLONY OPTIMIZATION APPROACH (FS-MACO) BASED FIVE LAYERED ARTIFICIAL NEURAL NETWORK FOR CROSS DOMAIN OPINION MINING

摘要

Web mining and web usage mining are attracting many researchers to propose new ideas, models, deploying machine learning algorithms and more. Internet usage expands its wings to almost all kind of applications which includes e-commerce. E-commerce facilitates the consumers/customers to buy the products online and at the same time, web analytics helps the website administrators to identify which products sell more. Opinion mining is the key to analytics in many decision-making tasks in the e-commerce arena. This research work aims to propose feature election using modified ant colony optimization approach (FS-MACO) based five layered artificial neural networks for cross-domain opinion mining. Dataset is obtained which consists of reviews about products such as books, DVDs, electronics and kitchen appliances. The features are identified by making use of modified ACO and opinion mining is performed by using ANN. Accuracy and F-measure are the metrics chosen for the evaluating the performance of the proposed work. Comparison of domain-specific and domain ? dependent words are presented. Results portray that the proposed work outperforms better than that of the existing work in terms of the chosen performance metrics.