data, while preserving these essential properties. By replacing the arithmetic mean
with more sophisticated, adaptable functions, by canceling out redundancies in the
inputs, one can improve the quality of automatic recommendations, and tailor recommender systems to specific domains.