Authors: Kachale Manali Bhagwan, Prof. C. A. Kolse, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Recommendation systems deliver personalized experiences on e-commerce, media streaming and social platforms by suggesting items that users are likely to value. Traditional collaborative filtering and content-based filtering face data sparsity and the cold-start problem for new users and items. This paper reviews AI-based recommendation with machine learning and deep learning. It explains neighbourhood-based and matrix-factorization collaborative filtering, content-based filtering and hybrid designs, then surveys deep learning recommenders including neural collaborative filtering, wide-and-deep and factorization machine models for click-through prediction, two-stage candidate generation and ranking at industrial scale, and self-attentive and bidirectional transformer models that learn from sequences of user actions. Evaluation with benchmark datasets and ranking metrics, and strategies for cold start, are discussed. A five-layer real-time recommendation architecture comprising data collection, feature processing, candidate generation, ranking and serving with feedback is proposed, along with research gaps, objectives, a comparison of approaches and challenges including popularity bias, filter bubbles, privacy and explainability. The review concludes that hybrid deep recommenders that combine behaviour, content and context deliver accurate real-time suggestions that improve user engagement.
