A recommender matrix-driven stacking SENet framework for predicting purchase decisions in E-commerce

Abstract Customer reviews can be quite significant for business outcomes and customer choices in the digital era. However, conventional systems have big problems as well as the increasing fake reviews as they are not so adept with resisting time limits and complexity in big amounts of data. User reviews play an important role in e-commerce product purchases, which is also not easy to be applied by recommender systems due to noisy feedback, counterfeit reviews and interactivity data. This research work suggests a stacking SENet model for the prediction of the purchase decision of e-commerce based on a recommender matrix. It is based on the stacking structure and feature refinement based on SENet and harmony search assistance of matrix factorization to enhance the reliability of recommendations. A clustering and prediction metrics were used to evaluate the method on the Amazon product review dataset. The experimental results demonstrate that the proposed approach provides better recommendation accuracy and minimum error rate as compared to state-of-the-art baseline approaches. This research work provides an integrated recommendation systems framework to enhance feature extraction, modeling of interactions, and the likelihood of a review being reliable, thus encouraging the practical deployment in the e-commerce industry.