Chair of Artificial Intelligence and Machine Learning
print


Breadcrumb Navigation


Content

Rank Aggregation for Incomplete Rankings (Ba/Ma)

Topic for a bachelor/master's thesis

Short Description:

Rank aggregation is aimed at finding a joint or consensus ranking among a set of preferences given over a set of alternatives. Considering the preference data as “noisy” observations of this consensus, one approach to tackle rank aggregation is via estimation of an underlying probabilistic model. In [1], a Generalized Method of Moments (GMM) is proposed for one of the most commonly used probabilistic models called Plackett-Luce (PL) model.

The goal of the thesis is to elaborate on extensions of this method, in particular the following ones: (i) Although most aggregation problems include incomplete rankings, the authors only consider the case of complete rankings (except for top-k and bottom-k observations, which correspond to special cases of incomplete rankings). (ii) The method is compared experimentally to the well-known Minorization-Maximization algorithm [2]. However, since the empirical study is quite limited in scope, this comparison should be expanded, especially through the use of real-world data sets [3].

Requirements

Addressing the aforementioned problems both theoretically and experi- mentally.

Prerequisites

Basic knowledge in machine learning and data analysis, programming skills.

Contact

Mohsen Ahmadi (ahmadim@mail.upb.de) or Prof. Eyke Hüllermeier

References

  • [1] Hossein Azari Soufiani, William Chen , David C. Parkes, Lirong Xia , Generalized Method-of-Moments for Rank Aggregation , Proceedings of the Annual Conference on Neural Information Processing Systems (NIPS 2013), Lake Tahoe, Nevada, USA.
  • [2] David R. Hunter. MM algorithms for generalized Bradley-Terry models. In The Annals of Statistics, volume 32, pages 384–406, 2004.
  • [3] Toshihiro Kamishima. Nantonac collaborative filtering: Recommendation based on order responses. In Proceedings of the Ninth International Conference on Knowledge Discovery and Data Mining (KDD), pages 583–588, Washington, DC, USA, 2003