Abstract
Generalized intent discovery aims to extend a closed-set in-domain intentclassifier to an open-world intent set including in-domain and out-of-domainintents. The key challenges lie in pseudo label disambiguation andrepresentation learning. Previous methods suffer from a coupling of pseudolabel disambiguation and representation learning, that is, the reliability ofpseudo labels relies on representation learning, and representation learning isrestricted by pseudo labels in turn. In this paper, we propose a decoupledprototype learning framework (DPL) to decouple pseudo label disambiguation andrepresentation learning. Specifically, we firstly introduce prototypicalcontrastive representation learning (PCL) to get discriminativerepresentations. And then we adopt a prototype-based label disambiguationmethod (PLD) to obtain pseudo labels. We theoretically prove that PCL and PLDwork in a collaborative fashion and facilitate pseudo label disambiguation.Experiments and analysis on three benchmark datasets show the effectiveness ofour method.