Trinity Fusion: A Deep Learning based Triad Collaborative Filtering System for Product Recommendations

Abstract

Recommender Systems (RS) enhance user experience by presenting relevant products based on their preferences and past interactions. Among various techniques, Collaborative Filtering (CF) is widely used due to its ability to generate personalized recommendations using the preferences of similar users. However, traditional CF methods, such as stochastic matrix factorization, model user–item relationships linearly and suffer from limitations including low learning efficiency, cold-start issues, and data sparsity. Traditional cooperative filtering is combined with advanced neural networks and terminology in order to solve these issues. “A Weighted Parallel Deep Triad Collaborative Filtering Model based on Singular Value Decomposition (SVD), Restricted Boltzmann Machine (RBM) and Ontology-based term weighting technique (OBTW)is proposed for significant improvement.”A user-item evaluation matrix is built from scratch. Singular Value Decomposition (SVD) is merged with the user-item matrix to provide a low-rank estimate of the matrices. It is followed by developing latent characteristics to anticipate consumer tastes by integrating the matrix of user items using RBM. OBTW consists of ontology development, weighting scheme and classifier module. Root Mean Squared Error(RMSE), Mean Absolute Error(MAE), Precision, Recall, F1-Measure and Normalized discounted cumulative gain (NDCG) are used to assess the accuracy of the proposed model on the Amazon Product Reviews dataset. The proposed model is compared with recent studies and existing collaborative filtering methods such as SVD, RBM and Probabilistic matrix factorization (PMF). It achieves lower RMSE (0.9443) and MAE (0.6589), along with higher Precision (93%), Recall (95%), and F1-score (93%), demonstrating improved recommendation accuracy and effectiveness.