Enhancing Rating Prediction Quality Through Improving the Accuracy of Detection of Shifts in Rating Practices

The most widely used similarity metrics in collaborative filtering, namely the Pearson Correlation and the Adjusted Cosine Similarity, adjust each individual rating by the mean of the ratings entered by the specific user, when computing similarities, due to the fact that users follow different rating practices, in the sense that some are stricter when rating items, while others are more lenient. However, a user’s rating practices change over time, i.e. a user could start as lenient and subsequently become stricter or vice versa; hence by relying on a single mean value per user, we fail to follow such shifts in users’ rating practices, leading to decreased rating prediction accuracy. In this work, we present a novel algorithm for calculating dynamic user averages, i.e. time-in-point averages that follow shifts in users’ rating practices, and exploit them in both user-user and item-item collaborative filtering implementations. The proposed algorithm has been found to introduce significant gains in rating prediction accuracy, and outperforms other dynamic average computation approaches that are presented in the literature.