A Utility Theoretic Approach to Determining Optimal Wait Times in Distributed Information Retrieval

This paper presents an approach for providing optimal access to multiple resources combined to a virtual resource through an information retrieval agent. The key design decisions of this agent are:

• source to use
• how long to wait for responses
• how to merge the results (ranking)

The main goal of this research is to determine the optimal termination time for the search (T).

Utility The authors use a similar utility model as Montgomery et. al but model the error term epsilon (representing all uncertainties associated with predicting utility) using a normal distribution with mean of 0 and a standard deviation 1.

Literature

• Kehoe et al. [10] show that consumers have significant costs associated with waiting.

Interesting Aspects The authors consider the potential decrease in utility due to overlapping results. The decision problem is formulated as finding the maximum of the expected value of the utility minus search cost. In the last section the authors present an adaptive heuristic, which decides even during the retrieval process based on the available data. They provide an ex-post analysis (with full knowledge of prices) and show that using the adaptive approach with this level of information yields better results than the static solution.