The Bradley-Terry-Luce (BTL) model is well-known in the literature on discrete choice (Bradley & Terry, 1952). It starts from the parametric model of pairwise comparisons:
Given n x n
matrix C
of pairwise comparisons, where C[i,j]
informs about how often the item x_i
wins over x_j
, the parameters
can be estimated by likelihood maximization:
the implementation of which is placed in BTL.py
. Check out this work for more details.