这篇论文反映的是佐治亚大学的一个研究小组在语义web组合方面的一个研究成果。这个小组在基于语义的服务发现，组合，调用和性能评估等方面都做了工作，这篇文章主要涉及其服务组合方面。其目录如下图所示
所谓语义，这篇文章主要使用了ontology，并且在不同的地方引入了对不同的ontology的引用。
文章的核心是一个semantic process template,在这个template中，可以指定此组合服务的组成成分服务（文中称为activity）及其各个成分服务之间的关联关系，显然这个template是描述用户process（flow）需求的（requirements）。下图就是一个semantic process template实例：
在指定各个activity的语义时，指定activity的功能（operation），输入，输出，在指定这几项时，用其所应用的ontology来进行限制。如下图所示
候选服务注册时，也注册相应的语义信息。则，discovery服务时，就会进行基于各种ontology的语义匹配。寻找合适的服务来完成各个activity。文章中具体论述为：“To perform semantic discovery according to the original implementation, the users can annotate the inputs and outputs of the service requirements and the discovery process in a UDDI will result in the services that match these semantic requirements. This discovery algorithm has been extended later in [14] that also supports annotating each operation in a WSDL file with a concept in functional ontology along with the annotation of preconditions and effects of that operation. Hence during discovery the service requirements are semantically annotated by associating it with concepts in ontologies that represent operation, inputs, outputs, preconditions and effects of the service. The discovery mechanism supported in MWSCF is based on all these kinds of annotations in addition to input and output semantics.”
另外一段文字说明了如何使用ontology进行语义匹配：“One of the outputs of the operation of interest is annotated with LSDISCargoServices: PickupDate and the other output is annotated with LSDISCargoServices:PickupLocationDetails which is a concept in the ontology that is 2 levels up in the hierarchy of subClassOf relationships. One of the outputs of the service represented by the operation in WSDL match exactly one of the outputs of the activity QuerySupplierPartner and the other output does not match exactly. For exact semantic matching the semantic matching value is 1 and for non-exact matches the semantic matching value is calculated using a linear function that decides the semantic matching value based on the subClassOf hierarchy. The final semantic matching value M3 is the average of semantic matching values of the two outputs. In the example the value of M3 is (1+0.8)/2 = 0.9. The subClassOf hierarchy and linear function used in this implementation can be extended with a better function that can be used to characterize two concepts separated in an ontology by a number of named relationship properties.”
解释一下，当template中规定的操作的输出与候选WSDL指定的候选服务的操作的输出在概念上完全一致时，则语义匹配值为1；当不完全一致时，就要根据概念之间的关系，用某种线性公式来计算概念的相似度了。
综合考虑三个因素，功能，输入，输出之间的语义相似度，下面的文字论述了如何计算template中的actrivity和WSDL中的候选服务之间的相似度：“The overall semantic matching value between the activity QuerySupplierPartner in the process template and the operation of interest in candidate WSDL, MS is the weighted arithmetic mean of the individual semantic parts, (1 * 50 + 1 * 30 + 0.9 * 20 ) / (50+30+20) = 0.98.”
最终的过程如图所示
文章中对自己工作的总结：1）对服务和过程（process）的富语义化；2）使用template实现可配置process；3）一个综合的服务组合框架；4）自适应性。尤其是对process指定语言的适应性。
现在不清楚的是，如何在服务注册机构中annotate与服务相关的操作operation，输入inputs，输出outputs，前件preconditions和效果effects。猜测会在另外一篇关于服务注册和发现的文章中具体论述。
METEOR-S Web Service Composition Framework (MWSCF)中包括四个部分：process builder， discovery infrastructure，XML repositories和process execution engine.