Two-Tier Enhanced Hybrid Deep Learning-Based Collaborative Filtering Recommendation System for Online Reviews

Abstract

Collaborative filtering-based recommender systems have recently attracted audiences due to the precise prediction of user interests and provide recommendations accordingly. The user-specific interests are the main requirement to build any recommendation model that produces the desired recommendation list. But the users' interests are sometimes unpredictable due to the fluctuating nature of the arrival of newer products. To resolve this problem and achieve better recommendation outcomes, a two-tier enhanced hybrid collaborative filtering based recommendation system (EHCFR) is constructed in this work based on deep learning. Initially, users in the dataset are segmented based on their age stratification to obtain users' interests based on age. Then, the major features are extracted from the dataset using the word learning enhanced variational auto-encoder (EVAE). These features are provided along with the rating matrix as the input to the deep belief network (DBN) for rating prediction. Based on the predicted ratings, the top N1 recommendation list is generated. Then, a time window strategy is adapted in the model to determine the dynamic fluctuations of user interests. Another list called the top N2 recommendation list is generated based on these fluctuations. Finally, both these lists are concatenated to provide accurate and favorable recommendations to the users. The proposed model is tested on the user dataset and provides competitive performance against the existing state-of-the-art techniques. Also, a reliable comparison is made with the existing popular datasets, such as Movielens 100k and Jester, and the results prove the efficacy of the proposed method.