Aggregation of Preferences in Recommender Systems

Aggregation of preferences, criteria or similarities happens at various stages in recommender systems. Typically such aggregation is done by using either the arithmetic mean or maximum/minimum functions. Many other aggregation functions which would deliver flexibility and adaptability towards more relevant recommendations are often overlooked. In this chapter we will review the basics of aggregation functions and their properties, and present the most important families, including generalised means, Choquet and Sugeno integrals, ordered weighted averaging, triangular norms and conorms, as well as bipolar aggregation functions. Such functions can model various interactions between the inputs, conjunctive, disjunctive and mixed behavior. Following, we present different methods of construction of aggregation functions, based either on analytical formulas, algorithms, or empirical data. We discuss how parameters of aggregation functions can be fitted to observed 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.