A Cost-Bounded Algorithm to Control Events Generalization

In order to build descriptions of prototypical situations, we first developed a system, MLK (Memorization for Learning Knowledge), allowing us to gather events related to similar situations starting from descriptions found in texts, these events being represented by conceptual graphs. One of the stages to build these prototypes consists of generalizing some similar graphs in order to produce a description. In this paper, we present a cost bounded algorithm of conceptual graph generalization, proceeding by ascending clustering. The use of costs on the operations of generalization allows us to control the growth of the search space.