The 2nd Computational Social Choice Competition


In conjunction with IJCAI 2024 in Jeju Island, South Korea


The field of computational social choice (COMSOC) combines ideas, techniques, and models from computer science and social choice theory for aggregating collective preferences. This thriving and multidisciplinary field of research has numerous applications to group decision-making, resource allocation, fair division, and election systems. One of the most well-studied problems in COMSOC focuses on designing voting mechanisms for selecting the winning candidates for an election. Paradoxes and impossibility results are commonly encountered when implementing voting rules in electoral systems. Researchers are therefore exploring alternatives to classical voting mechanisms by incorporating, for instance, principles and techniques from Machine Learning. Agent-based simulations can also tackle such challenges, as evidenced by their successful applications in negotiation research, supply chain management, and energy markets. In line with this vision, the Computational Social Choice Competition (COMPSOC) series capitalizes on the progress in agent research and computational social choice to drive the development of inclusive, robust, and fair election systems.

What is COMPSOC?

The 2nd Computational Social Choice Competition (IJCAI-COMPSOC 2024) aims to advance research in computational social choice by leveraging multiagent simulations and machine learning techniques. The competition will focus on the principled evaluation and analysis of voting rules in a competitive setting. The competitors will develop and submit the code of their voting rules, which will then be compared in a tournament based on social welfare and axiomatic satisfiability (anonymity, neutrality, monotonicity, Pareto optimality, unanimity, and non-imposition). The competition aims at providing valuable insights into the performances of voting mechanisms defined over parametrically generated voting problems, alternatives, and voters. COMPSOC will bring together researchers from computational social choice, social sciences, political sciences, multiagent systems, and machine learning and provide a unique benchmark for evaluating voting mechanisms in various synthetic (or real) problem domains. The competition also aims to advance the field by providing a systematic approach to designing and assessing voting mechanisms in the absence of established theoretical results. This advancement will help bridge the gap between axiomatic and experimental analysis of voting systems, ultimately leading to improved explainability.

Flow of the Competition

Your objective as a participant is to design rules that yield maximal social welfare while being robust against any variability in the profiles: number of voters, number of candidates, ballot distributions, and the possibility of having distorted preferences.  When optimizing the rules, you should consider the following (possibly unknown) parameters: TopN, distortion ratio, number of candidates, number of voters, etc.

The top four winning competitors are the competitors with the voting rules that yield the highest social welfare for the multiagent voters (given the baseline ballots of the competition) while satisfying the properties mentioned above. Various sample codes of well-known voting rules will be provided to the participants to guide their implementations (including Borda, Copeland, Dowdall, etc.).

In addition to submitting the Python code of their voting mechanisms, the participants are expected to submit a report describing their mechanism, implementation, and expected results. This will help disseminate the lessons learned from running the competition to the community and set the direction for future tournaments.

The detailed guidelines for the competitions can be found here.

For questions or requests, contact Rafik Hadfi (rafik.hadfi [at]