A Comparative Review of Convolutional Neural Networks, Long Short-Term Memory, and Recurrent Neural Networks in Recommendation Systems

Deep learning (DL) computing has emerged as the Gold Standard in the machine learning (ML) community in recent years. There are numerous recommender systems (RS) being used to treat information over-load problems in e-commerce, entertainment, and social media today. This review presents a comprehensive and comparative analysis of deep learning models based on Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNNs), Rectified Linear Unit-Deep Neural Network (ReLU-DNN), Adaptive Deep Learning-based method for the Recommendation System (ADRS) model, and several other hybrid DL methodologies in the context of RS. Open research problems in the domain, such as collection and feedback of user data, cold start, data sparsity, and scalability have been explored and studies to tackle these issues have been highlighted. It also assesses potential areas for future research and development of hybrid methods combining strengths of a multitude of architectures and incorporating attention mechanisms to enhance the recommendation quality metrics. We analyze the strengths and limitations of each architecture and find that hybrid techniques, like LSTM-RNN, demonstrated outstanding efficacy in problems of text classification. Overall, this review provides a comprehensive understanding of the applications and capabilities of CNNs, LSTM, and RNNs in recommendation systems, serving as a valuable resource for researchers and practitioners in the field. Further, the study will be comprehended by identifying an efficient model with prominent accuracy reports.