What is Rank Aggregation?

Rank aggregation is the process of combining multiple ranked lists into a single ranking. An example of a score-based rank aggregation method is shown below (picture from here):

Rank Aggregation in Recommendation Systems

As pointed out in Aggarwal, C.C. (2016). Advanced Topics in Recommender Systems (page 417), this is a relatively unexplored approach in the context of recommendation systems, where instead of a single algorithm, a certain set of algorithms is used that generate recommendations for a given user, and then the results of these algorithms are aggregated to create a new recommendation. Aggregation is not a trivial problem, as there is no single universal method for combining such rankings.

I have created a simple figure to demonstrate this process:

Where to start?

Blog Posts:

• Vlad Niculae - Kemeny Young Optimal Rank Aggregation in Python
• Daniel Alabi - Kemeny Rank Aggregation

Publications:

• Rank aggregation methods for the Web
• RankAggreg, an R package for weighted rank aggregation
• Is Rank Aggregation Effective in Recommender Systems? An Experimental Analysis

Software:

There is some excellent software written in Python and R which can help you.

Python

• Ranx - A Blazing-Fast Python Library for Ranking Evaluation, Comparison, and Fusion.
• Ranky - Another good library for ranking problem written in Python.
• RecRankAgg - Easy to use rank aggregation software for recommendation systems.
• MC4 - An implementation of Markov Chain Type 4 Rank Aggregation algorithm in Python.
• Kemeny_ranking - (Exact) Kemeny rank-aggregation using Integer-Programming with Python.
• Pyrankagg - Pure python code for rank aggregation/voting.

R

• RankAggreg - R package for weighted rank aggregation.

Java

• ERA - This repository contains the implementation of the Evolutionary Rank Aggregation (ERA) method proposed in here.

More:

• Repository about group recommender systems