The approach of this paper has been that a theoretical analysis of the possible effects of a variable, within the framework of the probabilistic theory of information retrieval, can give us useful insights, which can then be embodied in very much simpler formulae. This approach has shown very considerable benefits, enabling the development of effective weighting functions based on the three variables considered (term frequency within documents and queries and document length). These functions are simple extensions of the Robertson/Sparck Jones relevance weight, and therefore fit well with some other work on the development of probabilistic models.
The approach complements a number of other current approaches. In particular, it fits somewhere in between formal modelling and pure empiricism, alongside regression-based methods, to which it seems to offer some ideas.