Preference elicitation is a fundamental problem in single-unit combinatorial auctions, but it becomes prohibitive even for small instances of multi-unit combinatorial auctions. The bidders cannot express their preferences exactly as this would take a huge number of bids, typically leading to inefficient allocations.
Hence, markets with economies of scale and scope require more compact and yet expressive bidding languages. In this thesis, we propose an expressive bidding language allowing bidders to describe the characteristics of their cost functions. Bidders in these auctions can specify various discounts and markups, and specify pricing rules as logical functions. Finding the optimal allocation given these pricing rules is a strongly NP-hard optimization problem and we propose a mixed integer program to solve it.
Based on field data, we introduce a multi-item cost function and provide extensive computational experiments to explore the computational burden and the impact of different language features on the computational effort and total spend of the auctioneer. In addition, we explore characteristics of the knowledge representation of the bidding language.
«
Preference elicitation is a fundamental problem in single-unit combinatorial auctions, but it becomes prohibitive even for small instances of multi-unit combinatorial auctions. The bidders cannot express their preferences exactly as this would take a huge number of bids, typically leading to inefficient allocations.
Hence, markets with economies of scale and scope require more compact and yet expressive bidding languages. In this thesis, we propose an expressive bidding language allowing bid...
»