Avoiding Approximation Errors in Multi-Issue Negotiation with Issue Dependencies

Searching for good bids in a utility space based on multiple, dependent issues in general is intractable. Tractable algorithms do exist for independent issue sets, so one idea is to eliminate the dependencies by approximating the more complex utility space with issue dependencies. It has been shown that an approximation may give reasonable results when some structural features of the negotiation domain and preference profile are exploited. Of course, there is a risk that approximation results in significantly different negotiation outcomes. In this paper, we present a checking procedure to mitigate this risk and show that by tuning the parameters of this procedure the outcome deviation can be controlled. These parameters allow for a trade-off between computational cost and accuracy of negotiation outcome. Based on experimental results we propose specific values for the parameters of the checking procedure that provide a good balance between computational costs and accuracy. Additionally, we show how different values of these parameters influence the computational costs of negotiating multiple issues with dependencies.