基于软件网络的服务自动分类和推荐方法研究

摘要

服务的分类和推荐方法对于服务管理和组合都具有重要意义.文中利用服务组合历史,从面向服务软件的结构角度研究了服务分类和推荐方法:提出面向服务软件的软件网络模型,并用于抽象面向服务软件的拓扑结构；利用复杂网络社区发现算法对软件网络进行聚类,从而实现服务的自动分类；利用软件网络描述服务之间的可组合关系及其强度,并基于服务的使用场景提出了相应的服务推荐算法.使用ProgrammableWeb上API服务和mashup应用的真实数据来说明该文方法的有效性.从实验结果可以看出,该方法可以准确地实现服务分类,整个聚类的纯净度达到了86.8％,同时推荐的服务也比较有意义,这对于解决服务计算中的一些问题,包括服务的管理、检索、组合等都具有一定的意义.%Service classification and recommendation are two important problems in service management and composition. This paper, utilizing service composition histories, investigates the method for service classification and recommendation from a topological perspective of service-oriented software (SOS). It proposes a software network model for SOS to abstract its topological structures; it uses a community detection algorithm to cluster software networks and realizes service classification; it uses software networks to describe compositional strength between services and proposes an algorithm for service recommendation. API services and mashup applications in ProgrammableWeb are used as subjects to demonstrate the feasibility of the proposed approach. Experimental results show a high cluster purity of 86. 8%. Finally we discuss the implications of our approach on service computing.