Applying plan recognition algorithms to program understanding

Abstract

Program understanding is often viewed as the task of extracting plans and design goals from program source. As such, it is natural to try to apply standard AI plan recognition techniques to the program understanding problem. Yet program understanding researchers have quietly, but consistently, avoided the use of these plan recognition algorithms. This paper shows that treating program understanding as plan recognition is too simplistic and that traditional AI search algorithms for plan recognition are not suitable, as is, for program understanding. In particular, we show (1) that the program understanding task differs significantly from the typical general plan recognition task along several key dimensions, (2) that the program understanding task has particular properties that make it particularly amenable to constraint satisfaction techniques, and (3) that augmenting AI plan recognition algorithms with these techniques can lead to effective solutions for the program understanding problem.

Identification of design motifs with pattern matching algorithms

“…Most of these approaches use structural matching between micro-architectures and design motifs, with different algorithms being used: rule inference [5,22], queries [23,24], fuzzy reasoning nets [25], and constraint programming [6,26].…”

“…The main problem of such a structural approach is the inherent combinatorial complexity of identifying subsets of entities matching design motifs, which corresponds to a problem of subgraph isomorphism [27]. Approaches based on constraint programming [6] also face a combinatorial complexity, although explanations [28] reduce this complexity through userinteractions [26]. Antoniol et al introduced an approach to reduce the search space using metrics [7].…”

“…For example, some approaches used a Prolog-like unification mechanism [5] or constraint programming [6], which are slow because of the combinatorial explosion of possible occurrences, i.e. the possible combinations of entities in a system that form micro-architectures similar to a design motif.…”

Improving design-pattern identification: a new approach and an exploratory study

“…Queries have the potential to be extremely fast (Beyer et al 2005) but so far have been used only to specify motifs in a non-systematic way. Quilici et al (1997) used constraint programming to identify design motifs. Their approach consists of translating the problem of design-motif identification into a problem of constraint satisfaction.…”

Program understanding as constraint satisfaction

“…The plan and architecture recovery paradigms are very similar in problem representation-each exploiting both program constraints (structure) and architecture pattern constraints (knowledge) to reduce the computational expense of locating patterns in large source examples. • Quilici et al (1997b) investigate the similarities and differences between the Artificial Intelligence technologies and approaches to plan recognition in general with the more specific task of recognizing program plans and understanding legacy systems. This work demonstrated that the treatment of program understanding as plan recognition is too simplistic and that traditional AI search algorithms for plan recognition are not applicable, as is, to program understanding.…”