Pruning and aging for user histories in collaborative filtering

In this paper, we introduce algorithms for pruning and aging user ratings in collaborative filtering systems, based on their oldness, under the rationale that aged user ratings may not accurately reflect the current state of users regarding their preferences. The aging algorithm reduces the importance of aged ratings, while the pruning algorithm removes them from the database. The algorithms are evaluated against various types of datasets. The pruning algorithm has been found to present a number of advantages, namely (1) reducing the rating database size, (2) achieving better prediction generation times and (3) improving prediction quality by cutting off predictions with high error. The algorithm can be used in all rating databases that include a timestamp and has been proved to be effective in any type of dataset, from movies and music, to videogames and books.