Digital Business. First Iternational ICST Conference, DigiBiz 2009, London, UK, June 17-19, 2009, Revised Selected Papers

Research Article

A Recommender for Active Preference Estimate

Download
466 downloads
  • @INPROCEEDINGS{10.1007/978-3-642-11532-5_3,
        author={Andrea Malossini and Paolo Avesani},
        title={A Recommender for Active Preference Estimate},
        proceedings={Digital Business. First Iternational ICST Conference, DigiBiz 2009, London, UK, June 17-19, 2009, Revised Selected Papers},
        proceedings_a={DIGIBIZ},
        year={2012},
        month={5},
        keywords={},
        doi={10.1007/978-3-642-11532-5_3}
    }
    
  • Andrea Malossini
    Paolo Avesani
    Year: 2012
    A Recommender for Active Preference Estimate
    DIGIBIZ
    Springer
    DOI: 10.1007/978-3-642-11532-5_3
Andrea Malossini1, Paolo Avesani1
  • 1: Fondazione Bruno Kessler

Abstract

Recommender systems usually assume that the feedback from users is independent from the way a suggestion is presented. Usually the learning effort to estimate the user model is aimed at recognizing a subset of items according to the user preferences. We argue that acceptance or rejection is not independent from the presentation of the recommended items. Shaping a recommended item means to decide what kind of presentation can be more effective in order to increase the acceptance rate of the item suggested. In this work we address the challenge of designing an active policy that recommends a presentation of an item, i.e. which features use to describe the item, that will produce a more informative feedback from the user. The main contribution of this paper is a first principled design of an algorithm to actively estimate the optimal presentation of an item. An experimental evaluation provides the empirical evidence of the potential benefits.