A Novel Movie Recommendation System Based on Collaborative Filtering and Deep Learning
Sanaz Norouzi Larki · Artificial Intelligence Tools in Software and Data Engineering (AITSDE) · 2025
This study presents a hybrid recommendation system that integrates deep learning and association rules, aiming to enhance accuracy, long-tail item coverage, and explainability in sparse datasets. In the deep learning component, vector representations of users and items are learned using Multi-Layer Perceptrons (MLP) or interactive embedding models to capture non-linear relationships in user–movie interactions. In the association rule component, the efficient FP-Growth algorithm traverses the dataset twice to extract frequent itemsets and rules of the form based on support, confidence, and lift metrics. The outputs of both components are combined at the score-level fusion stage using a weighting factor α to calculate the final recommendation score for each user. The proposed method is evaluated on the MovieLens dataset and compared with baseline models including DL-Only, FP-Only, and traditional Collaborative Filtering (CF). Results show that the hybrid system achieves over 7% average improvement in Precision@K, Recall@K, MAP@K, and NDCG@K compared to the best individual models. Moreover, item coverage and performance stability across different α values are improved, with only a slight increase in computational cost. These findings demonstrate that combining neural representations with statistical rules can achieve a balanced trade-off among accuracy, efficiency, and explainability in recommender systems.