Generating k-best solutions to auction winner determination problems

Abstract

Auction participants cannot always articulate their requirements and preferences. Sometimes, for instance, the buyer in a procurement auction cannot quantify the value of non-price solution attributes or delineate between hard and soft constraints. This precludes formulating the winner determination problem (WDP) as an optimization problem. Existing decision-support aids for such situations extend an optimization framework. We present an approach that frames the decision problem as one of exploration rather than optimization. Our method relies on an algorithm that generates k -best solutions to auction WDPs. Our algorithm can incorporate hard constraints into the generation process and can scale to practical procurement auctions. We show how to extract useful guidance from k -best WDP solutions, and we evaluate our method using real bids submitted by real suppliers in an HP material parts procurement auction.