A Reinforcement Learning Approach to Achieve Unobtrusive and Interactive Recommendation Systems for Web-Based Communities

Abstract

Adaptive recommendation systems build a list of suggested links to nodes that usually cannot be reached directly from current web page. These recommendations are given by means of user models, where some parts of those models may be mined/learned from user’s interactions with a web site.

However, user’s interactions with the web site do not usually include user’s interaction with the recommendation system. In other words, most of current systems adapt recommendations to users just by “looking over her shoulder”. That occurs, in spite of the fact that taking into account user’s behavior upon recommendation should be a main part of the adaptation mechanism, because recommendation is not transparent to a user.

-Other recommendation systems interact with the user, but in an obtrusive way, making explicit requests to the user (prompting the user for rating) that are usually not followed. In this paper we present a recommendation system “in front of the user”, a system that looks directly at the user and interacts with her softly. Its key features are (i) it adapts to users by taking into account their interactions with a Web-based Communities Platform, and (ii) it adapts its own recommendations by unobtrusively taking into account the user behavior upon recommendations.