Rank Biased Overlap (RBO)

Rank Biased Overlap (RBO) is a measurement between two rankings that prioritizes overlap at the top of the ranking. This module provides a way to calculate RBO between two rankings, both as a standalone function and as an evaluation measure.

API Documentation

pyterrier_alpha.rbo(a, b, p=0.99)

Calculate the Rank Biased Overlap between two rankings. :rtype: Iterable[Tuple[str, float]]

Added in version 0.3.0.

Changed in version 0.12.5: Fixed bug where b wasn’t passed properly

class pyterrier_alpha.RBO(other, p=0.99, *, name=None)

Create an RBO measure from a dataframe of rankings. :rtype: Measure

Added in version 0.3.0.

Changed in version 0.3.1: Fixed bug where p wasn’t honored.

Acknowledgements

If you use this measure, be sure to cite:

Citation

Webber et al. A similarity measure for indefinite rankings. ACM Trans. Inf. Syst. 2010. [link]

@article{DBLP:journals/tois/WebberMZ10,
  author       = {William Webber and
                  Alistair Moffat and
                  Justin Zobel},
  title        = {A similarity measure for indefinite rankings},
  journal      = {{ACM} Trans. Inf. Syst.},
  volume       = {28},
  number       = {4},
  pages        = {20:1--20:38},
  year         = {2010},
  url          = {https://doi.org/10.1145/1852102.1852106},
  doi          = {10.1145/1852102.1852106},
  timestamp    = {Tue, 06 Nov 2018 12:51:56 +0100},
  biburl       = {https://dblp.org/rec/journals/tois/WebberMZ10.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

This implementation was based on the one provided by Charlie Clarke to ir-measures.