Trust- and Rating- based Recommendations for On-line Social Networks
Anna Kobusińska, Dionisis Margaris, Jakub Peikert · 2021
This paper analyzes the influence of incorporating trust rates between users, as well as objects’ ratings into the recommendation algorithms for bipartite on-line social networks. Three types of approaches are introduced: a trust-based approach that uses trust rates between users in the system to calculate the recommendation lists, a rating-based approach that uses exact ratings given to objects by users, and a trust-rating-aware approach that combines both of the mentioned above methods. The proposed approaches are introduced to probabilistic spreading and heat spreading recommendation algorithms. The first one strongly favors popular objects and focuses on accuracy metrics, but its diversity metrics are low. The latter one focuses on unpopular objects and diversity metrics but ignores accuracy metrics. Adding trust and object ratings to the recommendation algorithms changed a variety of their characteristics. Accuracy of results increased, and more personalized recommendations are obtained.