Best First Search Planning of Service Composition Using Incrementally Refined Context-Dependent Heuristics

Abstract

In oder to decide if a agent capability is helpful to achieve a goal, modern search algorithms in AI research use heuristics to narrow the search space by indicating which capability is the best to use. Considering the lack of information about pragmatic meaning, creating sound heuristics automatically out of capability descriptions asks too much of modern reasoning algorithms. Most approaches use semantics in oder to enable the reasoner to improve Word-sense disambiguation in their ontology matching tasks. As semantics are meant to be shared, the information is context independent and quite general. I postulate that context-dependent meaning can play an important role in describing the meaning of concepts used, as some meaning might change with the changes in context. The proposed thesis creates context-dependent heuristics by combining expert knowledge with machine learning. The PhD has the goal of structuring descriptions with a concept introduced in linguistics, introducing a description of domain knowledge and contextual information and thereby enable the automatic creation of context-dependent heuristics. Choosing from the many improvement points of agent planning, this work focuses on the improvement of capability descriptions.