Fuzzy-enhanced hybrid recommender system using autoencoder and GRU with fuzzy ACO weighted fusion

Personalization in large-scale systems is made possible by recommender systems. Collaborative filtering is a traditional approach for recommendation, but it struggles to handle inaccuracy, data sparsity, and prediction error. Our study proposes a hybrid recommendation model that integrates timestamps, user characteristics, and item characteristics to enhance prediction evaluation criteria, thereby addressing these limitations. A new user feature was created by fuzzy c-means clustering algorithm. Using three separate learning mechanisms, AutoEncoder (AE) and Singular Value Decomposition (SVD), the method produces initial rating predictions based on the user features and item features, which are combined with user-item ratings. The Gated Recurrent Unit (GRU) model takes a combination of user-item ratings and their timestamps as input to identify temporal patterns in behaviors and make another initial prediction. The independent predictions are combined by an optimization process, which is then followed by a weighted averaging step. The Fuzzy Ant Colony Optimization (Fuzzy ACO) method is utilized at this point to fine-tune the weights to carry out the ultimate prediction. The efficacy of our hybrid model was evaluated on the MovieLens 1M and MovieLens 100K datasets. In terms of RMSE, MAE, precision, and recall, our model surpasses baseline models in evaluation measures.