MARS: An Agent-Based Recommender System for the Semantic Web

Abstract

Agent-based Web recommender systems are applications capable to generate useful suggestions for visitors of Web sites. This task is generally carried out by exploiting the interaction between two agents, one that supports the human user and the other that manages the Web site. However, in the case of large agent communities and in presence of a high number of Web sites these tasks are often too heavy for the agents, even more if they run on devices having limited resources. In order to address this issue, we propose a new multi-agent architecture, called MARS, where each user’s device is provided with a device agent, that autonomously collects information about the local user’s behaviour. A single profile agent, associated with the user, periodically collects such information coming from the different user’s devices to construct a global user profile. In order to generate recommendations, the recommender agent autonomously pre-computes data provided by the profile agents. This recommendation process is performed with the contribution of a site agent which indicates the recommendations to device agents that visit the Web site. This way, the site agent has the only task of suitably presenting the site content. We performed an experimental campaign on real data that shows the system works more effectively and more efficiently than other well-known agent-based recommenders.

Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

This work has been partially supported by the MIUR–“Italian Ministry of Education, University and Research”, under the Research Project Quadrantis.