[Paper Review] Effectiveness of Preference Elicitation in Combinatorial Auctions

Benoît Hudson, Tüomas Sandholm|ArXiv.org|May 27, 2002

Auction Theory and Applications18 references4 citations

TL;DR

This paper evaluates preference elicitation in combinatorial auctions, proposing a novel bound-approximation query mechanism that drastically reduces information revelation by eliciting only approximate valuations initially and refining them only when needed. The method achieves near-optimal social welfare with a vanishingly small fraction of the bids required in traditional direct revelation mechanisms, even as the number of items and agents increases.

ABSTRACT

Combinatorial auctions where agents can bid on bundles of items are desirable because they allow the agents to express complementarity and substitutability between the items. However, expressing one's preferences can require bidding on all bundles. Selective incremental preference elicitation by the auctioneer was recently proposed to address this problem (Conen & Sandholm 2001), but the idea was not evaluated. In this paper we show, experimentally and theoretically, that automated elicitation provides a drastic benefit. In all of the elicitation schemes under study, as the number of items for sale increases, the amount of information elicited is a vanishing fraction of the information collected in traditional ``direct revelation mechanisms'' where bidders reveal all their valuation information. Most of the elicitation schemes also maintain the benefit as the number of agents increases. We develop more effective elicitation policies for existing query types. We also present a new query type that takes the incremental nature of elicitation to a new level by allowing agents to give approximate answers that are refined only on an as-needed basis. In the process, we present methods for evaluating different types of elicitation policies.

Motivation & Objective

• To address the high communication and computational cost of full preference revelation in combinatorial auctions where bidders must value all 2^k - 1 bundles.
• To evaluate and improve upon selective incremental preference elicitation as a scalable alternative to direct revelation mechanisms.
• To design and empirically validate new elicitation policies that minimize information disclosure while ensuring optimal social welfare.
• To introduce and assess a new query type—bound-approximation queries—that allows agents to provide rough bounds refined only when necessary.
• To theoretically and experimentally demonstrate that elicitation benefits scale with problem size, maintaining efficiency as items and agents increase.

Proposed method

• The auctioneer uses a constraint network to maintain lower and upper bounds on agent valuations for each bundle, propagating knowledge from prior queries.
• Queries are issued only when they cannot be inferred from previous answers, ensuring no redundant elicitation.
• A new query type—bound-approximation queries—allows agents to report approximate value bounds (e.g., lower and upper bounds), refined only if needed for optimal allocation.
• Elicitation policies are evaluated using a cost model where different query types have different costs, and performance is measured by total elicitation cost relative to full revelation.
• The system integrates order queries (comparing bundle values) and value queries (estimating bundle values), with hybrid policies combining both types.
• Theoretical analysis shows that unrestricted random elicitation minimizes revelation if savings are possible, and restricting elicitation to allocatable bundles is beneficial.

Experimental results

Research questions

• RQ1 Can selective incremental preference elicitation significantly reduce the amount of valuation information that must be revealed in combinatorial auctions?
• RQ2 How does the performance of elicitation policies scale with increasing numbers of items and agents?
• RQ3 What is the impact of using bound-approximation queries—where agents give rough value bounds refined only when necessary—on elicitation cost and efficiency?
• RQ4 Do hybrid elicitation policies combining bound-approximation and order queries outperform those using only one query type?
• RQ5 Under what conditions does restricting elicitation to allocatable bundles improve performance, and is this assumption theoretically justifiable?

Key findings

• In all studied elicitation schemes, the amount of information elicited is a vanishingly small fraction of that required in direct revelation mechanisms as the number of items increases.
• The bound-approximation query policy significantly reduces elicitation cost by deferring precise valuation until necessary, achieving high efficiency with minimal information disclosure.
• Hybrid policies combining bound-approximation and order queries outperform single-query-type policies, especially when order queries are relatively inexpensive.
• The method maintains its benefit as the number of agents increases, with elicitation cost growing sublinearly relative to full revelation.
• Experiments show that at 2 agents and 8 items, the mixed bound-approximation and order query policy averages 172 elicitation cost, compared to 230 for bound-approximation-only policies.
• Theoretical results confirm that if savings are possible, unrestricted random elicitation minimizes revelation, and restricting elicitation to allocatable bundles is beneficial.