Handling Granularity Differences in Knowledge Integration

Doo Soon Kim, Bruce Porter

Knowledge integration is a process of combining two different knowledge representations together. This task is important especially in learning where new information is combined with prior knowledge or in understanding where a coherent knowledge representation should be generated out of several knowledge fragments. A challenging problem in KI is handling granularity differences, i.e. combining together two knowledge representations with granularity differences. This paper presents an algorithm to find such correspondences between two representations with a granularity difference and to combine the two representations together based on the correspondences. The algorithm uses coarsening operators which generate coarse-grained representations from a representation. At the end, we introduce a large scale project in which the algorithm will be used.

Subjects: 10. Knowledge Acquisition; 12. Machine Learning and Discovery

Submitted: Sep 14, 2007