A Qos-Aware Web Service Recommendation System

Abstract

The adoption ofWeb Services is rapidly increasing on theWorldWideWeb, which is giving rise to the demand of an efficient Web Service quality evaluation system. Web service is a software based application providing standardized interface to enable interoperable communication between machines on the internet. Most of the Web services provide similar functionality; however, differ in nonfunctional characteristics i.e. response time or throughput. A recommender system intends to recommend an item to a user by predicting user’s expected rating on that particular item.In order to predict unknown rating value , RS uses the existing rating values and supporting information. In a QoS-Aware Recommender System, the quality of service is considered for recommending personalized web services to user. The QoS includes the response time and throughput that a user receives while invoking a web service. While there have been many researches carried out for recommending a specific web service to a user, there still remains room for development. Currently available methods only take into account the client-side QoS information, ignoring the services’ contextual attributes. Moreover, Web service QoS elements including response time and throughput, mainly depend on the location of the user relative to the Web service. The location factor is also rarely considered in the existing QoS prediction systems. The two new QoS-aware Web Services Recommendation Systems are proposed in this thesis, in which the contextual properties of a Web service, and the location of the user respective to the Web service are incorporated. Firstly, the proposed method process the WSDL files and collects the contextual properties from WSDL files in order to cluster Web services on the basis of their feature similarities. Thus, more accurate neighbor selection takes place and prediction value is determined using QoS record. A user influenced prediction value is also determined. In order to map both, service and user influence on QoS prediction, a hybrid memory-based CF model will be developed. A model-based CF model is also developed which incorporates Web services contextual characteristics and users location as well. First, users will be clustered depending on their locations. Secondly, the contextual properties will be collected from WSDL files, and then the web services will be clustered on the basis of similar features. After that, an advanced form of Matrix Factorization (MF) will be used to recommend user specific Web services. Large scale experimentation will be performed on WSDream datasets, which involve dif ferent service invocation records over 339 users. These experimental results will show how effective and reliable the proposed system is.