Recommender Systems as Dynamical Systems: Interactions with Viewers and Creators
Keywords: dynamical systems, recommendation, preferences, strategic behavior
TL;DR: A survey of papers modelling interactions between recommenders, viewer, and creators
Abstract: The design and evaluation of recommender systems often takes the perspective of supervised machine learning, treating viewer preferences and the content catalogue as static. However, in reality, recommender systems interact with and shape the behavior of viewers and content creators. In this position paper, we argue that due to these interactions, recommendation systems are more accurately characterized as dynamical systems, impacting the environment in which they operate. Towards this goal, we propose a unified framework of a recommender system as a dynamical system, and we formulate existing mathematical models of interactions between recommender systems, viewers, and creators from prior work within this framework. This framework allows us to identify the similarities and differences between these models, which we hope aids future development of mathematical models for recommender system dynamics.
Submission Number: 8
Loading