Bayesian Vote Elicitation for Group Recommendations

Maeva Caillat (1, 2) , Nicolas N. Darcel (3) , Cristina Manfredotti (1) , Paolo Viappiani (4)

1 MIA Paris-Saclay - Mathématiques et Informatique Appliquées
2 AgroParisTech
3 PNCA (UMR 0914) - Physiologie de la Nutrition et du Comportement Alimentaire
4 DECISION

Elicitation of preferences is a critical task in modern application of voting protocols such as group recommender systems. This paper introduces a Bayesian elicitation paradigm for social choice. The system maintains a probability distribution over the preferences (rankings) of the voters. At each step the system asks the question to one of the voters, and the distribution is conditioned on the response. We consider strategies to pick the next question based on value of information, conditional entropy, and a mix of these two notions. We develop this idea focusing on scoring rules and compare different elicitation strategies in the case of Borda rule.

Mots clés: preference elicitation; computational social choice; preference aggregation; Borda count

Domaines: Intelligence artificielle [cs.AI]; Applications [stat.AP]

Maeva Caillat, Nicolas N. Darcel, Cristina Manfredotti, Paolo Viappiani. Bayesian Vote Elicitation for Group Recommendations. From Multiple Criteria Decision Aid to Preference Learning (DA2PL 2020), Andrea Passerini (University of Trento); Vincent Mousseau (Centrale Supelec), Nov 2020, Trento, Italy. ⟨hal-03006639⟩