Personalized Service Recommendation Based on Pseudo Ratings by Merging Time and Tag Preference

Abstract

With the rapid development of service-oriented computing technologies, large amounts of Web services have been released on the Internet to facilitate system construction. Consequently, personalized service recommenders are emerging to handle information overload in service computing. Collaborative filtering (CF) is popular for service recommendations based on explicit ratings or Quality of Service (QoS) information provided by users. In real cases, however, it is difficult to collect explicit feedbacks since most available feedbacks are implicit. In this paper, the authors propose an implicit-to-explicit rating approach, which can leverage the implicit feedback from user's collecting behavior to build a CF-based recommender system for Web services. Firstly, a user-service binary matrix is constructed based on the collection records in the watchlist. Secondly, the reputation rating, publishing time and tag information of services are combined into the previous binary matrix to generate a pseudo rating matrix, which can reflect users' preference changing more precisely. Thirdly, both the traditional and enhanced user-based CF methods are used to generate a personalized service list with these pseudo ratings. Finally, a set of experiments are designed to validate the proposed service recommendation approach based on a large scale and real-world dataset from a wellknown service registry center Programmableweb (PWeb). The experimental results show that our hybrid recommendation method is more efficient and accurate than the traditional log-based CF methods.

Web service recommendation based on time‐aware users clustering and multi‐valued QoS prediction

“…Most of time‐adaptive approaches adopt collaborative filtering and address two types of time effects, which are the shifting of service bias and the shifting of user preferences …”

“…Memory‐based CF is used to exploit full data or a sample of it in order to generate a suitable forecast . This sub‐class is mainly characterized by its high effectiveness, in addition to its simplicity of implementation .…”

Time‐aware service recommendation: Taxonomy, review, and challenges

“…In TASR, the hybrid feedback could be applied in different ways. For example, authors extract implicit feedback and turn it to explicit feedback to exploit the extracted pseudo ratings . In addition, we could use two methods from the explicit and implicit feedback classes and compare their outputs to recommend the best items.…”

“…Other techniques are used to deal with the lack of explicit feedback, like the use of pseudo‐ratings in the work of Zhang et al…”

“…Moreover, the majority of the approaches suffer from data sparsity and cold‐start problems. This is due to the use of collaborative filtering like in the works of Yu and Huang and Zhang et al Other problems are related each to the used method or technique like ignoring the dependencies between QoS values in the work of Guidara et al As for the advantages, we can see that the majority of the approaches have improved the prediction accuracy and the quality of recommendation like in the works of Rana and Jain and Zhang et al However, not all of them have achieved this goal with the same success. Moreover, they were aiming to realize more efficient approaches by applying additional features like the consideration of timestamp.…”