Proceedings of the AAAI Spring Symposium on Machine Learning in Information Access, March 25-27, 1996., Stanford, California, USA
This paper presents an adaptive model for multi-agent learning based on the metaphor of economic markets that can cope with the non-stationary and partially observable nature of an information filtering task. Various learning and adaptation techniques - i.e. reinforcement learning, bidding price adjustment and relevance feedback - are integrated into the model. As a result of this integration learning through the model exploits market competition in order to dynamically construct mixtures of 'local experts' from selfish agents. The model is embedded into SIGMA (System of Information Gathering Market-based Agents) for information filtering of Usenet netnews. The functionality of the system is discussed together with work underway for its evaluation.