Skip to content

Latest commit

 

History

History
16 lines (11 loc) · 848 Bytes

README.md

File metadata and controls

16 lines (11 loc) · 848 Bytes

rank-centrality

Computes the Rank Centrality scores based on Negahban et al 2016 [1], given a list of pairwise comparisons.

Note it is assumed that the comparisons cannot result in a draw. If you want to include draws, then you can treat a draw between A and B as A winning over B AND B winning over A. So for a draw, you can add (A, B) and (B, A) to comparisons.

The regularized version is also implemented. This could be useful when the number of comparisons are small with respect to the number of unique items. Note that for properly ranking, number of samples should be in the order of n logn, where n is the number of unique items.

Resources

1 - Negahban, Sahand et al. “Rank Centrality: Ranking from Pairwise Comparisons.” Operations Research 65 (2017): 266-287. DOI: https://doi.org/10.1287/opre.2016.1534