Principles and Properties of a MAS Learning Algorithm: A Comparison with Standard Learning Algorithms Applied to Implicit Feedback Assessment

The purpose of this paper is to present a new learning algorithm based on an adaptive multi-agent system and to compare it with classical learning algorithms such as the Multi-Layer Perceptron (MLP), the Support Vector Machine (SVM), and the Decision Tree (DT). This comparison is made using data extracted from logs of a local citizen information search engine, called iSAC. It is based on the learning and the inference of the assessment of a real user with regard to the documents provided by iSAC in response to his request. The experimental evaluations show that our algorithm provides results at least as good as those achieved with classical learning approaches, in addition to its capability to function in dynamic and time constrained environments.