Learning plan libraries for case-based plan recognition

摘要

This paper addresses the indexing and retrieval issues in the context of the case-based plan recognition. The indexing and storage mechanisms utilize the knowledge about planning situations that enable the recognizer to focus its search to a subset of the plan library containing relevant past plans. A two-level abstract indexing scheme, along with the incremental construction of the plan libraries, may significantly reduce the retrieval efforts of the recognizer. Adding a third level of indexing may also improve the retrieval, but it may be computationally too expensive for some planning domains. Experimental results show the next action prediction accuracy with and without utilization of the two-level indexing scheme.