Unsupervised Information Extraction: Regularizing Discriminative Approaches with Relation Distribution Losses

Abstract : Unsupervised relation extraction aims at extracting relations between entities in text. Previous unsupervised approaches are either generative or discriminative. In a supervised setting, discriminative approaches, such as deep neural network classifiers, have demonstrated substantial improvement. However, these models are hard to train without supervision, and the currently proposed solutions are unstable. To overcome this limitation, we introduce a skewness loss which encourages the classifier to predict a relation with confidence given a sentence, and a distribution distance loss enforcing that all relations are predicted in average. These losses improve the performance of discriminative based models, and enable us to train deep neural networks satisfactorily, surpassing current state of the art on three different datasets.
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-02318233
Contributor : Étienne Simon <>
Submitted on : Wednesday, October 16, 2019 - 5:18:49 PM
Last modification on : Friday, October 18, 2019 - 11:46:51 AM

Licence


Distributed under a Creative Commons Attribution 4.0 International License

Links full text

Identifiers

Citation

Étienne Simon, Vincent Guigue, Benjamin Piwowarski. Unsupervised Information Extraction: Regularizing Discriminative Approaches with Relation Distribution Losses. ACL 2019 - 57th Annual Meeting of the Association for Computational Linguistics, Jul 2019, Florence, Italy. pp.1378-1387, ⟨10.18653/v1/P19-1133⟩. ⟨hal-02318233⟩

Share

Metrics

Record views

19