An Adaptive Filter for Preference Fine-Tuning in Recommender Systems

A recommender system may recommend certain items that the users would not prefer. This can be caused by either the imperfection of the recommender system or the change of user preferences. When those failed recommendations appear often in the system, the users may consider that the recommender system is not able to capture the user preference. This can result in abandoning to further use the recommender system. However, given the possible failed recommendations, most recommender systems will ignore the non-preferred recommendations. Therefore, this paper proposes failure recovery solution for recommender systems with an adaptive filter. On the one hand, the proposed solution can deal with the failed recommendations while keeping the user engagement. Additionally, it allows the recommender system to dynamically fine tune the preferred items and become a long-term application. Also, the adaptive filter can avoid the cost of constantly updating the recommender learning model.