Reputation based Buyer Strategies for Seller Selection in Electronic Markets

Abstract

Reputation based adaptive buying agents that reason about sellers for purchase decisions have been designed for B2C ecommerce markets. Previous research in the area of buyer agent strategies for choosing seller agents in ecommerce markets has focused on frequent purchases. In this thesis, we present reputation based strategies for buyer agents to choose seller agents in a decentralized multi agent based ecommerce markets for frequent as well as infrequent purchases. We consider a marketplace where the behavior of seller agents and buyer agents can vary, they can enter and leave the market any time, they may be dishonest, and quality of the product can be gauged after actually receiving the product. Buyer agents exchange seller agents' information, which is based on their own experiences, with other buyer agents in the market. However, there is no guarantee that when other buyer agents provide information, they are truthful or share similar opinions. First we present a method for buyer agent to model a seller agent's reputation. The buyer agent computes a seller agent's reputation based on its ability to meet its expectations of product quality and price as compared to its competitors. We show that a buying agent acting alone, utilizing our model of maintaining seller agents' reputation and buying strategy does better than buying agents acting alone employing strategies proposed previously by other researchers for frequent as well as for infrequent purchases. Next we present two methods for buyer agents to identify other trustworthy buyer agent friends who are honest and have similar opinions regarding seller agents, based on sharing of seller agents' information with each other. In the first method, buyer agent utilizes other buyer agents' opinions and ratings of seller agents to identify trustworthy buyer agent friends. Reputation of seller agents provided by trustworthy buyer agent friends is adjusted to account for the differences in the rating systems and combined with its own information on seller agents to choose high quality, low priced seller agent. In the second method, buyer agent only utilizes other buyer agents' opinions of seller agents to identify trustworthy buyer agent friends. Ratings are assigned to seller agents by the buyer agent based on trustworthy friend buyer agents' opinions and combined with its own rating on seller agents to choose a high quality, low priced seller agent to purchase from. We conducted experiments to show that both methods are successful in distinguishing between trustworthy buyer agent friends, whose opinions should be utilized in decision making, and untrustworthy buyer agent friends who are either dishonest, or have different opinions. We also show that buyer agents using our models of identifying trustworthy buyer agent friends have higher performance than a buyer agent acting alone for infrequent purchases and for increasing numbers of sellers in the market. Finally we analyze the performances of buyer agents with risk taking and conservative attitudes. A buyer agent with risk taking attitude considers a new seller agent as reputable initially and tends to purchase from a new seller agent if they are offering the lowest price among reputable seller agents. A buyer agent with conservative attitude is cautious in its approach and explores new seller agents at a rate proportional to the ratio of unexplored seller agents to the all the seller agents who have sent bids. Our results show that, when buyer agents are making decisions based on their own information, a buyer agent with conservative attitude has the best performance. When buyer agents are utilizing information provided by their trusted friends, a buyer agent with risk taking attitude and using only trusted friend buyer agents' opinions of seller agents has the best performance. In summary, the main contributions of this dissertation are: 1.A new reputation based way to model seller agents by buyer agents based on direct interactions. 2.A protocol to exchange reputation information about seller agents with other buyer agent friends based on the friends' direct interaction with seller agents. 3.Two methods of identifying trustworthy buyer agent friends who are honest and share similar opinions, and utilizing the information provided by them to maximize a buyer agent's chances of choosing a high quality, low priced seller agent to purchase from.