HABE: hybrid AdaBoost ensemble method of probabilistic matrix factorization recommendation system

Abstract

The PMF model is effective for addressing high-dimensional, large-scale, sparse, and imbalanced rating data, yet it may suffer from limitations in generalization and prediction accuracy in certain scenarios. To address these limitations, we propose a hybrid AdaBoost ensemble method within the PMF model. In this paper, we use two-stage algorithms in the model. Our approach uses a two-stage algorithm, whereby the first stage involves fuzzy clustering to calculate the scoring matrix of user-items, followed by neural network training to further enhance scoring prediction accuracy. The second stage involves using the rating matrix as the basis learner for training by different neural networks, and the final score prediction result is obtained through ensemble learning. Our proposed model was evaluated on the MovieLens and FilmTrust datasets, and its effectiveness was demonstrated. Due to its well-crafted architecture and robust representation learning capability, our model can be readily applied to various PMF model settings, such as PMF, FCM-PMF, Bagging-BP-PMF, and AdaBoost-SVM-PMF models. The experiments show that the mean absolute error(MAE) of the proposed method increases by 1.24% and 0.79% compared with the Bagging-BP-PMF model on two different datasets, and the root mean square error(RMSE) increases by 2.55% and 1.87%, respectively. Finally, our experiments show that our proposed approach performs well in various settings. By utilizing ensemble learning to train the weight of the base learner from different neural networks, our method improves the stability of score prediction. Additionally, our results verify the universality of our approach.