Structured Output Learning with Abstention: Application to Accurate Opinion Prediction - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2018

Structured Output Learning with Abstention: Application to Accurate Opinion Prediction

Résumé

Motivated by Supervised Opinion Analysis, we propose a novel framework devoted to Structured Output Learning with Abstention (SOLA). The structure prediction model is able to abstain from predicting some labels in the structured output at a cost chosen by the user in a flexible way. For that purpose, we decompose the problem into the learning of a pair of predictors, one devoted to structured abstention and the other, to struc-tured output prediction. To compare fully labeled training data with predictions potentially containing abstentions, we define a wide class of asymmetric abstention-aware losses. Learning is achieved by surrogate regression in an appropriate feature space while prediction with abstention is performed by solving a new pre-image problem. Thus, SOLA extends recent ideas about Struc-tured Output Prediction via surrogate problems and calibration theory and enjoys statistical guarantees on the resulting excess risk. Instantiated on a hierarchical abstention-aware loss, SOLA is shown to be relevant for fine-grained opinion mining and gives state-of-the-art results on this task. Moreover, the abstention-aware representations can be used to competitively predict user-review ratings based on a sentence-level opinion predic-tor.
Fichier principal
Vignette du fichier
garcia18a.pdf (895.44 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01950907 , version 1 (11-12-2018)

Identifiants

  • HAL Id : hal-01950907 , version 1

Citer

Alexandre Garcia, Slim Essid, Chloé Clavel, Florence d'Alché-Buc. Structured Output Learning with Abstention: Application to Accurate Opinion Prediction. 35th International Conference on Machine Learning18., Jul 2018, Stockholm, Sweden. ⟨hal-01950907⟩
71 Consultations
63 Téléchargements

Partager

Gmail Facebook X LinkedIn More