فهرست نشریات علمی دسترسی آزاد ایران

Recommendation systems are very useful in domains such as e-commerce, news portals and software requirement analysis. Collaborative filtering models have been used widely, these models often suffer from data-sparsity, interpretability and cold-start problems. To solve these issues, various machine-learning, deep-learning and kernel based models have been employed. Among these, trust based collaborative filtering and cross-domain recommendations have successfully solved the issues to some extent. However, in the recent literature, cross-domain recommendations (CDRs) are made by taking common user ratings. In our paper, we introduce a novel approach that combines CDRs with trust-aware collaborative filtering which employs ‘a partial item overlap’ scenario. Our model operates in two phases: an offline phase calculates trust between source and target users, coarse rating prediction using Adaptive Neuro-Fuzzy Inference System (ANFIS), and clustering via Firefly Fuzzy C-Means (FAFCM); and an online phase, where cluster information and item similarities are used to generate personalized recommendations. Evaluated on the Douban and Movielens datasets using MAE and RMSE metrics, our approach demonstrates improved performance compared to existing methods, effectively mitigating common limitations in recommendation systems.