Bayesian Brain meets Bayesian Recommender - Towards Systems with Empathy for the Human Nature
This work addresses the challenge of modeling human behavior in recommender systems for better user experience, but it is incremental as it integrates existing theories without major breakthroughs.
The paper tackles the problem of noisy user feedback in recommender systems by applying the Bayesian brain theory from cognitive neuroscience, showing through real user experiments that multicomponent user models improve prediction of human behavior.
In this paper we consider the modern theory of the Bayesian brain from cognitive neurosciences in the light of recommender systems and expose potentials for our community. In particular, we elaborate on noisy user feedback and the thus resulting multicomponent user models, which have indeed a biological origin. In real user experiments we observe the impact of both factors directly in a repeated rating task along with recommendation. As a consequence, this contribution supports the plausibility of contemporary theories of mind in the context of recommender systems and can be understood as a solicitation to integrate ideas of cognitive neurosciences into our systems in order to further improve the prediction of human behaviour.