Abstract
We drew on revised Reinforcement Sensitivity Theory to claim that users with an anxiety-related behavioral inhibition would experience proactively delivered recommendations as potential threats. Such users would display higher user engagement especially when they were interrupted by inaccurate (vs. accurate) recommendations, because they ruminate about them. This prediction was tested and confirmed in a controlled experiment that exposed participants to proactive recommendations on their smartphone. Results highlight the need to gain more knowledge on the neural correlates of anxiety, and to apply such insights to human–computer interaction design for recommender systems. © 2018, Springer International Publishing AG.