The seminal 2003 paper by Cosley, Lab, Albert, Konstan, and Reidl, demonstrated the susceptibility of recommender systems to rating biases. To facilitate browsing and selection, almost all recommender systems display average ratings before accepting ratings from users which has been shown to bias ratings. This effect is called Social In Bias (SIB); the tendency to conform to the perceived “norm” in a community. We propose a methodology to 1) learn, 2) analyze, and 3) mitigate the effect of SIB in recommender systems. In the Learning phase, we build a baseline dataset by allowing users to rate twice: before and after seeing the average rating. In the Analysis phase, we apply a new nonparametric significance test based on the Wilcoxon statistic to test whether the data is consistent with SIB. If significant, we propose a Mitigation phase using polynomial regression and the Bayesian Information Criterion (BIC) to predict unbiased ratings. We evaluate our approach on a dataset of 9390 ratings from the California Report Card (CRC), a rating-based system designed to encourage political engagement. We found statistically significant evidence of SIB. Mitigating models were able to predict changed ratings with a normalized RMSE of 12.8% and reduce bias by 76.3%. The CRC, our data, and experimental code are available at:

http://californiareportcard.org/data/