Adapting Ratings in Memory-Based Collaborative Filtering using Linear Regression

We show that the standard memory-based collaborative filtering rating prediction algorithm using the Pearson correlation can be improved by adapting user ratings using linear regression. We compare several variants of the memory-based prediction algorithm with and without adapting the ratings. We show that in two well-known publicly available rating datasets, the mean absolute error and the root mean squared error are reduced by as much as 20% in all variants of the algorithm tested.