Analysis And Community Detection On Web Services Network

Service-orientated computing(SOC) has been receiving considerable attention fromboth academia and industry, which focuses on publishing, registering, organizing andmanaging the huge amount of web services, and satisfying end-users’ requirementsfrom service discovery/composition. However, advances of Internet and newcomputing paradigms impose new challenges upon SOC. First, as web services arecontinued growing in size and scope, they are forming a services-centric complexsystem on the Internet. Second, services currently are short of semantics, and cannotsense each other, which pose “service silos”. To address these issues, this thesisfocuses on semantic sensing/connecting and community detecting for web services.This thesis is aiming at facilitating SOC from a crossing-research byincorporating Semantic Web, complex network and topic modeling(TM). Byincorporating Semantic Web, automatic SOC can be achieved by building semanticservices, in which services can be interpreted and automatically processed bymachines. However, traditional semantic services were essentially derived fromannotating WSDLs via external knowledge (e.g., ontology), which could suffer fromimprecise semantics and the adaptability problems of external ontology. Towards suchan issue, we enrich the semantics of services from the text description by leveragingTM. In addition, complex network analysis is used for managing and analyzing thestructures, laws and communities for the services-centric complex system.Specifically, we make the following contributions:1. Analyzing and optimizing web service-based networkAs web service-based network captures the underlying interoperability amongservices and pre-computes potential composing patterns, efficient composing can beachieved by applying searching algorithms to service-based network. We model thesemantic services into two novel “bottom-up” networks, and perform enrichingservice-based network on the issues of isolated services from network analysis bypredicting valuable services, and present a mechanism of pruning negative links fromreusing users’ historical composite services.2. Detecting communities for networked web servicesWe present mining service communities for capturing the insight dynamics forweb service-based network, which can be useful in organizing and managing web services system, and benefiting service discovery/composition by efficiently andaccurately locating demanded services. Herein we focus on two service communities:competition-oriented community structure from the functional semantics of services,and collaboration-oriented community structure computed from the interoperabilityamong services. We also present the usage of the two community structures inrefining users’ requirements and designing efficient service composing.3. Network analysis and mining integrating patterns of web APIs/Mashups.By modeling APIs/Mashups with various “top-down” network models, wepresent a comprehensive analysis on the functional and collaborative features forAPIs/Mashups, and analyze the small-world effect, power-law distribution and cliqueson the constructed networks. Specially, we also present the practical uses that arisefrom above findings, such as refining users’ requirements interactively and in a pilotway.4. Topic modeling for SOCTo address the imprecision and domain-sensitive problems on using externalontology in semantic annotating, we discuss the significance of combining TM inSOC. We perform clustering classical WSDL services from rich description byleveraging LDA, and present how TM benefits detecting service communities. Inaddition, we exploit the possibility of conducting TM on multi-labeled APIs byincorporating ontology with Labeled LDA. Experiments conducted on real-world APIdata show that the proposed method outperforms Labeled LDA, and demonstrate thesignifcance of incorporating ontology for refining redundant topics.