Article Preview
TopIntroduction
Web services became vital for the business of many companies in the software industry, especially with the advent of the software on demand paradigm, such as SaaS (Software as a Service). As in any business, providers have interest in understanding the needs of their clients to avoid customer attrition, and to attract new clients. Many providers focus on large scale service provision, and have very little knowledge about their clients. At the same time, they face hard decisions related to the maintenance of deployed services, service versioning to avoid breaking clients, and service redesign evolution to keep up with clients expectations. Typically, these decisions are made without a clear understanding of the possible outcomes, frequently based on worst-case scenarios. Understanding the usage clients make of services is thus invaluable to support web service life-cycle (Papazoglou, Andrikopoulos et al., 2011).
Data mining techniques have been applied in many business segments to discover knowledge about clients, which is hidden in large volumes of data (Tan, Steinbach et al., 2006). Web service mining (Liang, Chung, et al., 2006) aims at discovering patterns of service usage, i.e. specific ways in which web services (or their operations) are used repeatedly by a group of users with similar properties, as well as are correlated to each other. Usage analysis have been used to support the recommendation of services (Yu, 2012; Zhang, Ding et al., 2011; Kang, Liu et al., 2012; Rong, Liu et al., 2009), the discovery of service composition communities (Zhang, Yin et al., 2009; Wang, Wang et al., 2012), or process discovery for applications such as process documentation, conformance checking or process optimization (Motahari-Nezhad, Saint-Paul et al., 2011; Musaraj, Yoshida et al., 2010; Tang & Zou 2010; van der Aalst, 2012). van der Aalst (2012) highlights that, even when predefined interaction models are available, very often the reality differs of the expected behavior, justifying the deployment of sophisticated techniques to capture the actual usage patterns of services by their client applications.
Our work is focused on the usage analysis as a support for the service evolution life-cycle (Yamashita, Vollino et al., 2012; Silva, Vollino et al., 2012; Yamashita, Becker et al., 2012). Our approach is to empower providers with an understanding of the overall impact of changes in the whole set of client applications, enabling sound decisions in terms of evolution strategies. Providers can leverage usage impact information to make decisions about the creation, maintenance and decommissioning of versions. For that purpose, they must have a clear understanding of the patterns involved in the overall requests clients make (the operations they request, the structure of the messages exchanged, co-occurrence of operations, among others), and leverage these patterns to group clients with a similar service usage behavior, which we refer to as usage profiles.