Ranking Web Services with Limited and Noisy Information

With the increasing popularity of web services on the Internet, besides functionalities, Quality of Service (QoS) is becoming an important concern for describing characteristics of web services. QoS rankings provide valuable information for making optimal service selection and recommendation from a set of functionally similar or equivalent service candidates. However, in order to obtain such rankings, a huge number of invocations on the services are usually required, which is extremely expensive and even impractical in reality. To tackle this challenge, this paper proposes a scheme to derive the global ranking from observations of QoS rankings on subsets of the services, while the observations may also be contaminated by noise and errors. We introduce a pairwise comparison model to describe the relationships between services, and thus the ranking can be formulated as random walks over the services. A Markov chain based approach is proposed, and algorithms for deriving global rankings are designed. The efficacy of our approach is validated by both mathematical analysis and simulation experiments.