Behavior-based information for improving priority-rule search

Determine whether using behavioral-distance information alongside fitness for candidate screening, final-rule selection, reference-rule selection, and diversity management can improve the search efficiency or quality of reusable priority rules for the resource-constrained project scheduling problem.

Background

The paper finds a strong positive association between behavioral distance from a repeatedly rediscovered reference rule and test-set performance: rules behaviorally closer to the reference tend to achieve lower average deviation from the best-known lower bound. The authors suggest that behavioral similarity could support candidate screening, final-rule selection, reference selection, and diversity maintenance, but the experiments do not incorporate this information directly into the LLM-guided search.

The unresolved issue is whether behavior-based information provides benefits beyond fitness-based evaluation when integrated into the search process. In particular, it remains to be determined whether behavioral-distance criteria can reduce search effort, improve the quality of the final reusable priority rule, or improve population and archive management.