Skip to main content

Über dieses Buch

This volume contains ten thoroughly refereed and revised papers detailing recent advances in research on designing trading agents and mechanisms for agent-mediated e-commerce. They were originally presented at the 13th International Workshop on Agent-Mediated Electronic Commerce (AMEC 2011), collocated with AAMAS 2011 in Taipei, Taiwan, or at the 2011 Workshop on Trading Agent Design and Analysis (TADA 2011), collocated with IJCAI 2011 in Barcelona, Spain.

The papers presented at these two workshops illustrate both the depth and broad range of research topics in this field. They range from providing solutions to open theoretical problems in online scheduling and bargaining under uncertainty, to designing bidding agents in a wide area of application areas, such as electronic commerce, supply chain management, or keyword advertising, to designing agents that can successfully replicate actual human behaviors in realistic games.



Non–cooperative Bargaining with Arbitrary One–Sided Uncertainty

Non-cooperative bargaining is modeled as an extensive–form game with uncertain information and infinite actions. Its resolution is a long–standing open problem and no algorithm addressing uncertainty over multiple parameters is known. We provide an algorithm to solve bargaining with any kind of one–sided uncertainty. Our algorithm reduces a bargaining problem to a finite game, solves this last game, and then maps its strategies with the original continuous game. Computational complexity is polynomial with two types, while with more types the problem is hard and only small settings can be solved in exact way.
Sofia Ceppi, Nicola Gatti, Claudio Iuliano

An Adaptive Proportional Value-per-Click Agent for Bidding in Ad Auctions

Sponsored search auctions constitutes the most important source of revenue for search engine companies, offering new opportunities for advertisers. The Trading Agent Competition (TAC) Ad Auctions tournament is one of the first attempts to study the competition among advertisers for their placement in sponsored positions along with organic search engine results. In this paper, we describe agent Mertacor, a simulation-based game theoretic agent coupled with on-line learning techniques to optimize its behavior that successfully competed in the 2010 tournament. In addition, we evaluate different facets of our agent to draw conclusions about certain aspects of its strategy.
Kyriakos C. Chatzidimitriou, Lampros C. Stavrogiannis, Andreas L. Symeonidis, Pericles A. Mitkas

Improving Prediction in TAC SCM by Integrating Multivariate and Temporal Aspects via PLS Regression

We address the construction of a prediction model from data available in a complex environment. We first present a data extraction method that is able to leverage information contained in the movements of all variables in recent observations. This improved data extraction is then used with a common multivariate regression technique: Partial Least Squares (PLS) regression. We experimentally validate this combined data extraction and modeling with data from a competitive multi-agent supply chain setting, the Trading Agent Competition for Supply Chain Management (TAC SCM). Our method achieves competitive (and often superior) performance compared to the state-of-the-art domain-specific prediction techniques used in the 2008 Prediction Challenge competition.
William Groves, Maria Gini

Testing Adaptive Expectations Models of a Continuous Double Auction Market against Empirical Facts

It is well known that empirical financial time series data exhibit long memory phenomena: the behaviour of the market at various times in the past continues to exert an influence in the present. One explanation for these phenomena is that they result from a process of social learning in which poorly performing agents switch their strategy to that of other agents who appear to be more successful. We test this explanation using an agent-based model and we find that the stability of the model is directly related to the dynamics of the learning process; models in which learning converges to a stationary steady state fail to produce realistic time series data. In contrast, models in which learning leads to dynamic switching behaviour in the steady state are able to reproduce the long memory phenomena. We demonstrate that a model which incorporates contrarian trading strategies results in more dynamic behaviour in steady state, and hence is able to produce more realistic results.
Neil Rayner, Steve Phelps, Nick Constantinou

Autonomously Revising Knowledge-Based Recommendations through Item and User Information

Recommender systems are now an integral part of many e-commerce websites, providing people relevant products they should consider purchasing. To date, many types of recommender systems have been proposed, with major categories belonging to item-based, user-based (collaborative) or knowledge-based algorithms. In this paper, we present a hybrid system that combines a knowledge based (KB) recommendation approach with a learning component that constantly assesses and updates the system’s recommendations based on a collaborative and item based components. This combination facilitated creating a commercial system that was originally deployed as a KB system with only limited user data, but grew into a progressively more accurate system by using accumulated user data to augment the KB weights through item based and collaborative elements. This paper details the algorithms used to create the hybrid recommender, and details its initial pilot in recommending alternative products in an online shopping environment.
Avi Rosenfeld, Aviad Levy, Asher Yoskovitz

A Bidding Agent for Advertisement Auctions: An Overview of the CrocodileAgent 2010

Sponsored search is a popular form of targeted online advertising and the most profitable online advertising revenue format. Online publishers use different formats of unit price auctions to sell advertising slots. In the Trading Agent Competition Ad Auctions (TAC/AA) game, intelligent software agents represent a publisher which conduct keyword auctions and advertisers which participate in those auctions. The publisher is designed by game creators while advertisers are designed by game entrants. Advertisers bid for the placement of their ads on the publisher’s web page and the main challenge placed before them is how to determine the right amount they should bid for a certain keyword. In this paper, we present the CrocodileAgent, our entry in the 2010 TAC AA Tournament. The agent’s architecture is presented and a series of controlled experiments are discussed.
Irena Siranovic, Tomislav Cavka, Ana Petric, Vedran Podobnik

Dealing with Trust and Reputation in Unreliable Multi-agent Trading Environments

In shared competitive environments, where information comes from various sources, agents may interact with each other in a competitive manner in order to achieve their individual goals. Numerous research efforts exist, attempting to define protocols, rules and interfaces for agents to abide by and ensure trustworthy exchange of information. Auction environments and e-commerce platforms are such paradigms, where trust and reputation are vital factors determining agent strategy. And though the process is always secured with a number of safeguards, there is always the issue of unreliability. In this context, the Agent Reputation and Trust (ART) testbed has provided researchers with the ability to test different trust and reputation strategies, in various types of trust/reputation environments. Current work attempts to identify the most viable trust and reputation models stated in the literature, while it further elaborates on the issue by proposing a robust trust and reputation mechanism. This mechanism is incorporated in our agent, HerculAgent, and tested in a variety of environments against the top performing agents of the ART competition. The paper provides a thorough analysis of ART, presents HerculAgent s architecture and dis-cuss its performance.
Iraklis Tsekourakis, Andreas L. Symeonidis

Analysis of Stable Prices in Non-Decreasing Sponsored Search Auction

Most critical challenge of applying generalized second price (GSP) idea in multi-round sponsored search auction (SSA) is to prevent revenue loss for search engine provider (SEP). In this paper, we propose non-decreasing Sponsored Search Auction (NDSSA) to guarantee SEP’s revenue. Each advertiser’s bid increment is restricted by minimum increase price (MIP) in NDSSA. The MIP determination strategy influences bid convergence speed and SEP’s revenue. Fixed MIP strategy and Additive-Increase/Multiplicative-Decrease (AIMD) principle are applied to determine MIP values, and they are evaluated in this paper. For the convergence speed analysis, fixed MIP strategy converges faster than AIME in most instances. For SEP’s revenue, AIMD assists SEP to gain more revenue than fixed MIP strategy by experiments. Simultaneously, SEP’s revenue in Vickrey-Clarke-Groves auction (VCG) is the lower bound of that in AIMD.
ChenKun Tsung, HannJang Ho, SingLing Lee

Acceptance Strategies for Maximizing Agent Profits in Online Scheduling

In the global logistics market, agents need to decide upon whether to accept jobs offered sequentially. For each offer, an agent makes an immediate selection decision with little knowledge about future jobs; the goal is to maximize the profit. We study this online decision problem of acceptance of unit length jobs with time constraints, which involves online scheduling. We present theoretically optimal acceptance strategies for a fundamental case, and develop heuristic strategies in combination with an evolutionary algorithm for more general and complex cases. We show experimentally that in the fundamental case the performance of heuristic solutions is almost the same as that of theoretical solutions. In various settings, we compare the results achieved by our online solutions to those generated by the optimal offline solutions; the average-case performance ratios are about 1.1. We also analyze the impact of the ratio between the number of slots and the number of jobs on the difficulty of decisions and the performance of our solutions.
Mengxiao Wu, Mathijs de Weerdt, Han La Poutré


Weitere Informationen

Premium Partner