Considering neighbor projection on neural based recommender systems

Abstract

Now, CF is applied with a neural network to make the model more flexible and more accurate. Different neighbors have a different influence on the target user, and different users usually have different rating patterns. Therefore, the proposed method needs to consider two major issues when applying CF with a neural network: the similarity levels between the neighbors and the target user and the user's rating pattern conversion. Thus, the proposed method consists of three main modules to solve the issues mentioned above: rating conversion, similarity module uses, and prediction module. In the experiment, the proposed method is evaluated and compared with the current neural CF with friends and latent factor model on two types of datasets: real-world and synthetic datasets. In real-world datasets, N neighbors and all neighbors are evaluated to demonstrate the significance of the number of neighbors. Furthermore, the rating conversion module's performance is assessed by comparing the results of the proposed method with and without the rating conversion module. For the synthetic datasets, this work simulates the full rating matrix datasets and the partial rating matrix dataset to compare the effectiveness of using different types of distribution and dataset size. The experimental results demonstrate that the proposed method effectively outperformed the baselines utilizing ranking evaluation and prediction accuracy on real-world and synthetic datasets.