Generalized fractional strategy for recommender systems with chaotic ratings behavior

To provide useful and accurate recommendations, the role of recommender systems for e-commerce industry is to predict users' interest by approximating users' preferences and items characteristics. The increasing role of recommender systems in the e-commerce industry requires efficient recommendations to deal with the chaos in the users' ratings patterns. In this regard, different fractional gradient based adaptive algorithms are employed through matrix factorization to improve the efficiency of recommender systems in terms of recommendations speed and accuracy. In this paper, the generalized variant of fractional stochastic gradient descent termed as GFSGD is exploited for recommender systems. The proposed GFSGD improves the performance of the recommender systems with the capability to enhance the memory effect by capturing the ratings history more effectively. The proposed GFSGD assures fast convergence speed for higher fractional order values while, taking fractional order unity reduces the GFSGD to the standard SGD. The correctness of the GFSGD is confirmed through various evaluation metrics (RMSE, MAE, sMAPE, NSE), learning rates, fractional orders, and latent features. While the exactness of the proposed scheme is also verified through Movie-Lens and FilmTrust datasets.