A Recommender Network Perspective on the Informational Value of Critics and Crowds

Tune in for the live stream on YouTube.

Abstract: How do the ratings of critics and amateurs compare and how could they be combined? Previous research has generated variable results on the first sub-question, whereas the second remains unanswered. We have created a new dataset, with wine ratings from both critics and amateurs, and simulated the behavior and performance of the k-nearest-neighbor algorithm. We then formalized the social network spanned by k-nn and studied people’s relative influence. We find that critics are more consistent and predictable when compared to amateurs, and thus drawing advice from critics outperforms drawing advice from amateur crowds. Getting advice from both groups can further boost performance by a small margin. We then used our network-theoretic approach to identify influential critics, talented amateurs and the information flow between the groups. Our results provide evidence on the informational function of critics, and our methods can be leveraged to design better and more transparent recommender systems.