Plans as structured networks of hierarchically and temporally related case pieces

Abstract

Abstract. This paper describes a representation of plan cases as a structured set of goals and actions. These goals and actions are the unit pieces that form a case. These case pieces are related each other by hierarchical and temporal links (explanations) forming a tree-like network. We give importance not just to explicit links, i.e., links between case pieces which are concretely known, but also to implicit ones, i.e., possibly unknown links between case pieces. Each case piece is explained by antecedent links and explains other case pieces by consequent links. The retrieval of a case piece is mainly guided by its links and by its surrounding case pieces. Our concept of case piece usefulness is briefly explained. We discuss the benefit of reusing and directly accessing small case pieces from multiple cases for improving the Case-Based Reasoning (CBR) systems' capability and efficiency to solve problems. We explain the importance of stepwise refinement in plan cases and also the role that temporal representation can take in the meaningful and coherent construction of planning problem solutions. An application in musical composition domain is presented. We also show how a musical composition task can be treated as a planning task.

Experimental study of a similarity metric for retrieving pieces from structured plan cases: Its role in the originality of plan case solutions

“…Some systems like CELIA (Redmond, 1990), JULIA (Kolodner, 1989), PRODIGY/ANALOGY (Veloso, 1992), etc., break up the goal into smaller sub-goals, enabling plan construction by composition of sub-plans. This leads to a hierarchical representation of plan cases (Macedo et al, 1996a). The case representation is similar to a tree where each node is a goal and its sons the sub-goals, or at the latest level, the actions of the plan.…”

A structured framework for representing time in a generative composition system

“…The choice of cases for change and adaptation is made using a specific metric function [8] that weights several aspects of the chosen case or piece of case (snippet, [10]), according to the domain's characteristics and user's choice.…”

“…For the kind of tasks we deal with, it is vital to evaluate features like internal context (e.g., notes in a phrase), external context (hierarchy, structure) and temporal position. In our system, this evaluation is made according to a similarity metric developed by Macedo et al [8],where our temporal representation assumes two main roles: first, it simplifies the access to the external context; second, a node address is an important attribute to take into account when comparing nodes. The latter part demands a more detailed explanation.…”