Semantic-based Cloud Services Discovery and Recommendation

Abstract

Substantial number of vendors are offering their applications as Cloud Services (CS) to leverage the benefits of Cloud Computing (CC). Businesses and individuals have to select the CS that suits their requirements from a large pool of services. Consequently, there is an increasing demand for enhancing such a time-consuming and error-prone task. This thesis proposes CSRDS, a semantic-based system that allows the registration, discovery and selection of cloud services. The main building blocks of the proposed system include: Hybrid matchmaking of CSs, metadata semantic similarity model, semantic-based business-oriented request-service matchmaking algorithm and CS selection based on non-functional requirements. Despite its efficiency, the proposed CSRDS system provides the same set of services to all users, with no consideration of users interest or history. Motivated by the fact that the cloud users appreciate customized recommendations, we propose CSRec, a personalized reputation-based QoS-aware recommender system for CSs. Its main building blocks include: Feedback-based CS evaluation, hybrid collaborative filtering approach for personalized CS recommendation, new users similarity measure, dynamically built user profile, reputation calculation for CSs and 3-D justification for the recommended CSs. The key enabler to the realization of the proposed work is the proposed CS ontology, a comprehensive full-fledged CS domain ontology with querying capabilities. It serves as a knowledge base for the proposed systems. Extensive experimental evaluation is conducted in order to assess our proposed work. In order to carry out reliable evaluation, three real-world service datasets are used. The evaluation metrics include matchmaking, prediction and set recommendation metrics. The proposed work is compared to benchmark approaches. Case studies demonstrate that the proposed system is feasible, effective and relevant. Experimental results show that the proposed system has superior performance against the traditional approaches in real use-cases. The performance enhancement is investigated under different parameter settings. Besides, the evaluation of the introduced CS ontology consists of three processes. First, its coverage, consistency and expressiveness were verified via semantic reasoner and semantic modeling. Second, its feasibility was confirmed via prototype implementation while its practicability was validated via case study. Finally, the ontology quality evaluation metrics prove that CS ontology captures a comprehensive, consistent, rich, accurate and relevant representation of the CS domain and that its concepts are interpretable.