Cognitive Dissonance can Increase Consumption: On the effects of Inaccurate Explanations on User Behavior in Video-On-Demand Platforms

On most video-on-demand platforms, recommender systems help users find relevant content. Many recommender systems can provide explanations (e.g., short texts) on why they make recommendations for an individual user. Existing literature assumes that explanations are accurate, meaning that they align with the recommended content. Thus, the effect of inaccurate explanations that do not fit the displayed content is unclear. In this large-scale, real-world randomized controlled trial, users of a major Public Service Media provider received recommendations either without explanations (baseline), with accurate explanations (control), or with some inaccurate explanations (treatment). For users who experience only a low percentage of inaccurate explanations, usage (click-through rate) is similar to the case where all explanations are accurate; for users who experience a higher percentage of inaccurate explanations, usage significantly increases (up to +7.8%). For the extreme case, where most explanations are inaccurate, usage significantly decreases (down to -15.8%). The increase in usage (click-through rate) when experiencing some inaccurate explanations can be explained by Festinger's theory of cognitive dissonance. It postulates that users confronted with inconsistent cognitions try to remove dissonant cognitions by investigating the reason for a dissonant recommendation through clicking on it. As expected, usage decreases if most of the shown explanations are inaccurate. The findings demonstrate the benefits of online user studies, which allow for deeper insights into the interplay between recommender systems and human users. Furthermore, such results that may seem at first counter-intuitive for the recommender system community can be explained when drawing on well-established theories explaining human behavior.