22.6 Conclusions
The purpose of this chapter has been to present the state of the art in aggregation
functions and introduce established families of these functions that have properties useful for the purposes of recommendation. This has included means defined
with various weights, Choquet integrals defined with respect to fuzzy measures, tnorms/t-conorms which can be built from generators, and representable uninorms.
Many of the current methods used in recommender systems involve constructing
weighted arithmetic means where weights are determined by varying measures of
similarity, however in many cases the accuracy and flexibility of functions could
be improved with only slight increases to complexity. We have provided a number of illustrative examples of the different ways in which aggregation functions
can be applied to recommendation processes including ratings aggregation, feature