Software agents and multi-agent systems have been recognized as a new paradigm to
design and develop open distributed systems around loosely coupled autonomous
software components. These autonomous agents interact with each other asynchronously
to provide the benefits of robustness and localization while addressing the complexity
in real-world systems (Jennings and Wooldridge, 1998).
Interaction is fundamental to multi-agent systems. It can vary from simple
information exchange to managing interdependencies among agents. Interaction can
be either cooperative or competitive. In cooperative interaction, the agents may share
a common goal and work as a team to achieve it. On the other hand, agents in a
competitive interaction have individual goals which may conflict with each other.
These agents engage in a negotiation process and resolve conflicts to pursue their
own interests. Negotiation is essential in many applications such as e-commerce and
sharing of resources. Through negotiation, self-interested agents come to a mutually
acceptable agreement on conflicting issues. It provides satisfaction to the participating
agents (Jennings et al., 2001).
Negotiation process varies in duration and complexity depending on the context.
In a multi-agent e-commerce, buyer and seller agents negotiate intelligently on behalf
of their human counterparts. Considering the complexity of human negotiation process,
automated negotiation has been a major research challenge.
There are three primary approaches to automated negotiation found in multiagent
literature: game-theoretic (Rosenschein and Zlotkin, 1994), heuristic (Faratin
et al., 2001) and argumentation-based (Parson et al., 1998). As claimed in Jennings
et al. (2001), game-theoretic and heuristic-based approaches have many limitations.
In game-theoretic and heuristic approaches, agents’ preferences are fixed, complete
and correct. Agents cannot influence other agents’ preferences or internal mental
attitudes. Both the approaches allow agents to exchange proposals but do not allow
the agents to express any meta-information. Agents in these approaches have complete
information. They know the space of possible deals and also know how to evaluate
such deals.
|