Time Aware Implicit Social Influence Estimation to Enhance Recommender Systems Performances

Nowadays, e-commerce websites like Amazon, streaming platforms like Netflix and YouTube, and social networks like Facebook and Instagram play a significant role in our daily lives. However, with the constantly growing addition of items on these platforms, it becomes challenging for users to select the products that interest them. Hence, the implementation of recommender systems to facilitate this selection process. To enhance these recommender systems, some studies integrate social influences through trust and friendship information among users to whom recommendations are intended. However, this process of estimating social influence does not consider time and is based on explicits relationships of trust between users, which is not reassuring since these informations are not always available on e-commerce sites. In this paper, we propose to incorporate the temporal aspect into the process of estimating social influences, but using implicit trust informations (ratings that users give to items), which is much more available. Specifically, we have modified the basic recommender systems by incorporating the results of time aware social influence estimations based on implicit trust (through ratings of users). For our experiments, we used the epinions and ciao datasets, which are two platforms where users provide reviews on products from various domains. These experiments demonstrate that considering the temporal aspect of social influence effectively contribute to the improvement of these recommender systems performances. More precisely, we obtained an improvement from 0.902 to 0.833 following the MAE (Mean Absolute Error) metric and from 1.179 to 1.071 following the RMSE (Root Mean Square Error) metric for the Epinions dataset, and for the Ciao dataset, we obtained an improvement from 0.738 to 0.687 following MAE and from 1.023 to 0.953 following RMSE metric.