The families of aggregation functions defined in Section 22.3.2 are convenient to
use when trying to understand and interpret the results. The weights and parameters
have a tangible meaning and fitting these functions essentially involves finding the
best values for each parameter to maximize the reliability of the RS.
In other situations however, the interpretation side of things may not be as important: we just want to predict the unknown ratings reliably and automatically. There
are many non-parametric methods for building aggregation functions, which do not
have the advantage of system interpretation, however can be constructed automatically and fit the data closely. One “black-box” type method is to build a general
aggregation operator piecewise from the data. We can ensure that monotonicity and
boundary conditions are specified by smoothing the data and ensuring these properties hold for each individual segment. We consider here, the construction of spline
based aggregation functions [10].