A Study of Query Generation Strategies for Interactive Constraint Acquisition

摘要

The work presented in this paper is concerned with interactive constraint acquisition. In this paper, we have compared a number of different heuristics for query generation for interactive constraint acquisition. We have demonstrated that the choice of generalization strategy is closely related to the degree to which the user is being helpful and version space topology. Future work will address the utility of hybrid strategies for query generation for constraint acquisition. In addition, we will focus on learning disjunctive constraints and also, more importantly, sets of constraints. While in this paper we have compared a helpful teacher with one which is more helpful, a further investigation focused on how the teachers ability to provide good examples can affect the acquisition process needs to be undertaken. Such a study would more formally study the interaction between an intelligent learner with teachers of varying abilities and, in particular, study whether skill is more important than helpfulness.