Do IR models satisfy the TDC Retrieval Constraint?

Stéphane Clinchant (1) , Éric Gaussier (2)

1 LIG - Laboratoire d'Informatique de Grenoble
2 AMA - Analyse de données, Modélisation et Apprentissage automatique [Grenoble]

If it has been show in previous studies (as [5, 2]) that most IR models satisfy most IR constraints, the situation of the TDC constraint is unclear, and the goal of this short paper is to show that several state-of-the-art IR models indeed do not comply with the general TDC constraint, but do satisfy the speTDC one. We will review here the recently intro- duced log-logistic model [2], as well as the Jelinek-Mercer and Dirichlet language models.

Mots clés: axiomatic constraint; TDC constraint

Domaines: Apprentissage [cs.LG]; Intelligence artificielle [cs.AI]

Stéphane Clinchant, Éric Gaussier. Do IR models satisfy the TDC Retrieval Constraint?. 34th Annual ACM SIGIR Conference, Jul 2011, Beijing, China. pp.1155-1156, ⟨10.1145/2009916.2010096⟩. ⟨hal-00742614⟩