Continuous Double Auction (CDA) is an efficient market institution for real-world trading. Negotiation capabilities for software agents are a central concern. Specially, agents need to be able to prepare bids for and evaluate offers on behalf of the users they represent with the aim of obtaining the maximum benefit for their users. They do this according to some bidding strategies. However, in many cases, on one hand, determining which strategy to employ is a complex decision making task because of the inherent uncertainty and dynamics of the auction market; on the other hand, strategies in the literature do not adapt very well to the dynamic markets. To this end, this thesis is concerned with developing novel bidding strategies for CDAs and enhancing the performance of different strategies in CDAs with respect to adaptivity by designing some generally used tools. / In this thesis, we focus on two types of CDAs. One is the CDAs with a deadline of inactive interval. Another is the CDAs with a fixed deadline. Three kinds of adaptive behaviors are proposed to enhance the performance of the widely adopted strategies in CDAs in the literature. They are adaptive softness, adaptive judgment of price acceptability, and adaptive time strategies. First, in the CDAs with a deadline of inactive interval, we design novel adaptive strategies, named Adaptive Attitude strategies, based on eagerness. Eagerness indicates the current supply and demand relationship from the agent's own point of view. To compute the value of eagerness, fuzzy sets and fuzzy logic are used to cope with the significant degrees of uncertainty in CDA markets. We define two kinds of adaptive behaviors, adaptive softness and adaptive judgment of price acceptability. Both of them resemble human traders' behaviors to compromise and set thresholds on acceptable prices in the trading process of real life markets and can enhance the performance of various strategies. Secondly, in the CDAs with a fixed deadline, time strategies are researched by us. In this market, every agent is aware of the time. Therefore adaptive time strategies are introduced to guide the agent to arrange his behavior according to time, which can enhance the performance of different strategies. Both the novel strategies and the enhanced strategies have been demonstrated to be superior in a wide range of CDA circumstances. We show that eagerness is a, practical solution for this class of application. We believe that this work represents an important step towards adapting agents in auctions. / Through the work in this thesis, Adaptive Attitude (AA) strategies have been demonstrated to be superior in a wide range of CDA scenarios. Moreover, three kinds of adaptive behaviors have been shown to greatly enhance the performance of the widely adopted strategies in CDAs. / Ma Huiye. / "August 2006." / Adviser: Ho Fung Leung. / Source: Dissertation Abstracts International, Volume: 68-03, Section: B, page: 1731. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2006. / Includes bibliographical references (p. 151-162). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts in English and Chinese. / School code: 1307.

Sui, Xin. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2009. / Includes bibliographical references (p. 91-97). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- MAS-Based Resource Allocation Problems --- p.1 / Chapter 1.2 --- Combinatorial Auction As a Solution --- p.4 / Chapter 1.3 --- Strategy Issues --- p.5 / Chapter 1.4 --- Structure of This Work --- p.7 / Chapter 2 --- Combinatorial Auctions and Bidding Strategies --- p.9 / Chapter 2.1 --- Combinatorial Auctions for Resource Allocation --- p.9 / Chapter 2.2 --- Bidding Strategies in Combinatorial Auctions --- p.11 / Chapter 2.2.1 --- Berhault's Strategies --- p.11 / Chapter 2.2.2 --- An's Strategies --- p.13 / Chapter 2.2.3 --- Schwind´ةs Strategies --- p.15 / Chapter 2.2.4 --- Wilenius´ةs Strategy --- p.17 / Chapter 2.2.5 --- Overview of Previous Strategies --- p.18 / Chapter 3 --- An Adaptive Bidding Strategy in Static Markets --- p.21 / Chapter 3.1 --- Basic Concepts --- p.22 / Chapter 3.2 --- The Core Algorithm --- p.24 / Chapter 3.3 --- Experimental Evaluation --- p.31 / Chapter 3.3.1 --- Experiment Setup --- p.32 / Chapter 3.3.2 --- Experiment Results and Analysis --- p.33 / Chapter 4 --- An Adaptive Bidding Strategy in Dynamic Markets --- p.38 / Chapter 4.1 --- Basic Concepts --- p.39 / Chapter 4.2 --- The Core Algorithm --- p.42 / Chapter 4.3 --- Experimental Evaluation --- p.48 / Chapter 4.3.1 --- Experiment Setup --- p.49 / Chapter 4.3.2 --- Experiment Results and Analysis --- p.51 / Chapter 5 --- A Q-Learning Based Adaptive Bidding Strategy in Static Mar-kets --- p.59 / Chapter 5.1 --- An Overview of Q-Learning --- p.60 / Chapter 5.2 --- Basic Concepts --- p.63 / Chapter 5.3 --- The Core Algorithm --- p.65 / Chapter 5.4 --- Experimental Evaluation --- p.70 / Chapter 5.4.1 --- Experiment Setup --- p.70 / Chapter 5.4.2 --- Experiment Results and Analysis --- p.72 / Chapter 6 --- Discussion --- p.82 / Chapter 6.1 --- Applicability of the Adaptive Strategies --- p.82 / Chapter 6.2 --- Generalization of the Adaptive Strategies --- p.83 / Chapter 7 --- Conclusion and Future Work --- p.86 / Chapter 7.1 --- Conclusion --- p.86 / Chapter 7.2 --- Future Work --- p.88 / Bibliography --- p.91

Recent years have seen extensive studies on the pricing problem, as well as its many variances. They have found important applications in computational economics. Nowadays typical applications can be found in internet advertising, Google’s Auction for TV ads and many other resource allocation problems in electronic markets. In electronic markets, thousands of trading activities are processed in the internet or done automatically by computer programs. It is highly required that the trading mechanisms are efficient enough. In the thesis, we will study various pricing problems from different perspectives. The first problem we study is the design of auction mechanism when bidders are unit-demand. It can be applied in internet advertising. Thousand of advertisers bid for space in webpages to show their advertisements. We model the new problem and apply the General Second Price (GSP) mechanism to the problem. GSP is an efficient mechanism with linear time complexity. Moreover, we show that GSP has an envy-free equilibrium which can maximize the profit of advertisers. Auction mechanisms where bidders can bid for multiple items are also studied. A famous example of such auction is the Dutch flower auction. Such multi-unit auctions are widely studied these years. But budget constraints are not considered in many previous works. We study the scenario that each bidder has a budget on the money paid to the auctioneer and the valuation functions of bidders are non-linear. For the model, we design an adaptive clinching auction mechanism. The mechanism is proved to be incentive-compatible, which encourages bidders to reveal their true values, and Pareto-optimal, which ensures that no bidder can improve her utility without decreasing those of others. In some auctions, the items on sale are not available at the same time. For example, TV stations sell time-slots for advertisements on a daily basis. The advertisers are arriving and departing online and bidding for a set of timeslots. For the auction, we design a competitive mechanism which is truthful, i.e., all bidders have the incentive to submit their true private values to the auctioneer. Another important property the mechanism achieves is promptness, which makes sure that any advertiser that wins some time-slots could learn her payment immediately after winning these time-slots. In some pricing problems, upon the arrival of a new buyer, the seller needs to decide immediately whether he will sell his goods or not and what is the price. When buyers are unit-demand and each seller has b items on sale, the online pricing problem can be modelled by online weighted b-matching problem. For the problem, we show a randomized algorithm which achieves near-optimal competitive ratio. When buyers are not unit-demand, things are much more complicated. We consider a general model in which each buyer wants to buy a bundle of items and has a non-increasing valuation function for those items. We design a randomized algorithm which achieves low competitive ratio and derive a non-trivial lower bound on the competitive ratios. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy