A service recommendation approach based on trusted user profiles and an enhanced similarity measure

Numerous services issued from the emergence of web technologies drive research on how to provide users with trusted and credible services aligned with their needs. To tackle the service targeting problem, recommender systems have been developed. They are grouped into content-based approaches and collaborative filtering based approaches. Strongly focused on the target user profile, content-based methods are inaccurate when the target user profile is poor. To remedy this, collaborative filtering based methods exploit past experiences from many users. In the literature, they are organized into rating methods and ranking methods. In this paper, we propose a trusted collaborative filtering based approach that combines assets of both rating and ranking methods. Our proposal is built on an enhanced hybrid similarity measure and a novel trustworthiness score that is thereafter used to select trusted and relevant user profiles involved in the prediction process. By employing a customized ranking measure, our method improves the service ranking precision without affecting the rating prediction accuracy. Experiments are conducted on the WS-Dream dataset containing 339 users and real-world Quality of Service values related to 5825 web services. Compared to state-of-the-art collaborative filtering based methods, the obtained results show that our proposal offers the best trade-off in terms of rating prediction accuracy and ranking prediction accuracy.