CF recommenders that ask for explicit ratings information will usually do so on a finite ordinal scale - e.g. {1 = didn’t like it!,..., 5 = loved it!}. On
the other hand, it may be possible to convert user actions into ordinal values
as part of their profile - e.g. {regularly views, sometimes views, etc.}. Where
there is an ordinal scale, these values can be turned into numbers and aggregated.
For non-homogeneous functions and those which lack the shift-invariance property, it will be necessary to express these ordinal values on the unit interval. The
coarseness of the aggregated values may make the difference between, say, the
weighted arithmetic mean and the geometric mean negligible. Examples of aggregation functions particularly suitable for the aggregation of ordinal data are
the Sugeno integral and the induced OWA.