Constraint-directed negotiation of resource reallocations

The resource reallocation problem requires multi-agent choices under multiple-criteria, most of which are based on qualitative attributes. The conditional specification of a reallocation request (e.g., requiring a swap for one workstation with another) results in chains of reallocation transactions, which increase in complexity with the number of resources and agents. Also, the initial intentions for transactions may differ from the final transaction due to give-or-take on resource components. This work is motivated by the human negotiation procedures, such as logrolling, bridging, and unlinking. We view the process of reallocation negotiations as being constraint based. Constraints can be used both for evaluation of existing alternatives as well as for creating new ones. We define a set of qualitative evaluation and relaxation (alternative generation) techniques based on human negotiation problem solving. The search uses several aspects of constraints, such as constraint importance, looseness, utility and threshold levels. We evolve a mixed problem solving approach in which agents search individually in the early stages and as a group in the later stages. The constraint-directed negotiation approach is validated for the quality of solution in comparison to expert human negotiators on a variety of negotiation problems using a partial factorial design. The final version of the problem solver performs marginally better than the expert on experimental problems.