In general, a semantic service composition problem is to compose a set of semantic services from a given set of semantic services into one composite service that fulfills the given requirements. Any solution to this problem, in contrast to those for non-semanitc web service composition, can in principle make use of formal ontology-based service annotations.
The majority of current tools for automated semantic service composition is adopting means for logic-based AI planning off-line, or on-line in stochastic domains for this purpose. That requires an appropriate conversion of the semantic model of services and domain of the service composition problem into the corresponding formal planning domain description such as in PDDL. Other approaches perform a graph-based semantic service composition without AI planning techniques. In particular, they generate a service composition graph based on dynamically computed semantic service I/O parameter matching relations only, and extract the final composition plan from such a graph which can be optimized with respect to, for example, plan length or QoS. One challenge of optimal QoS-based and semantic service composition is to effectively integrate constrained combinatorial optimization with logic-based semantic composition planning. Main characteristics of semi-automated approaches for pattern-based or model-driven composition of semantic services are required user interactions at some or all stages of the composition process, or pre-given workflow patterns that are to be instantiated with services. Semantic service search and selection tools can be integrated with automated or semi-automated composition tools in order to heuristically prune the search space of composition and make suggestions to the user at each composition step, but they are inherently not suitable to perform the composition themselves. Finally, in compliance with the semantic SOA paradigm, any generated composite semantic service is assumed to be grounded in an executable orchestration of those web services it has been semantically composed of.
Current approaches for automated semantic service composition can be classified as static or dynamic, decentralized or centralized, and as functional- or process-level composition.
Functional- and process-level composition. Functional-level composition (FLC) of semantic services only exploits the functional (IOPE) service semantics described in given semantic service pro les, while process-level composition (PLC) also takes the semantic process models, if available, into account for accomplishing the given goal or service request.
Static Vs. dynamic composition. In static composition the generation and execution of semantic service plans is strictly decoupled. This holds for the majority of current approaches for automated semantic service composition which exploit classical AI planning methods under the assumption of a closed, perfect world with deterministic actions and complete initial state of a fully observable domain at design time. In fact, the problem of static service composition can be converted into a static state-based planning problem (I, A, G) in AI to devise a sound, complete, and executable plan which satisfies a given goal state G by executing a sequence of services as actions in A from a given initial world state I. On the other hand, dynamic composition approaches allow for planning under uncertainty about state changes and the outcomes of actions at planning time. We can further differentiate between restricted dynamic, advanced dynamic, and reactive dynamic composition. Restricted dynamic service planners permit the execution of information gathering under IRP (Invocation and Reasonable Persistence) assumption but not world state altering services during planning. Advanced dynamic composition planners are capable of heuristic replanning subject to partially observed but not caused world state changes that affect the current plan at planning time. In both cases the interleaved execution of planning with world state altering services is prohibited to prevent obvious problems of planning inconsistencies and conflicts. Reactive service composition planning of semantic services is fully interleaving service plan generation with execution in the fashion of dynamic contingency and real-time action planning in AI.
See Semantic Services. Composition of manufacturing process services in the cloud.