Declarative Belief Set Merging Using Merging Plans

摘要

We present a declarative framework for belief set merging tasks over (possibly heterogeneous) knowledge bases, where belief sets are sets of literals. The framework is designed generically for flexible deployment to a range of applications, and allows to specify complex merging tasks in tree-structured merging plans, whose leaves are the possible belief sets of the knowledge bases that are processed using merging operators. A prototype is implemented in MELD (MErging Library for Dlvhex) on top of the dlvhex system for HEX-programs, which are nonmonotonic logic programs with access to external sources. Plans in the task description language allow to formulate different conflict resolution strategies, and by shared object libraries, the user may also develop and integrate her own merging operators. MELD supports rapid prototyping of merging tasks, providing a computational backbone such that users can focus on operator optimization and evaluation, and on experimenting with merging strategies; this is particularly useful if a best merging operator or strategy is not known. Example applications are combining multiple decision diagrams (e.g., in biomedicine), judgment aggregation in social choice theory, and ontology merging.