Skip to Main content Skip to Navigation
New interface
Conference papers

Modèles en Caractères pour la Détection de Polarité dans les Tweets

Abstract : Character-level Models for Polarity Detection in Tweets We present our contribution to the DEFT 2018 shared task, with three entries based on different methods to perform topic classification and polarity detection for tweets in French, to which we added a voting system. Our first entry is based on lexicons (for words and emojis), character n-grams and a classifier implemented with a support vector machine (SVM), while the other two are endogenous methods based on character-level feature extraction : first a long short-memory recurrent neural network (BiLSTM) feeding a classifier implementing a multi-layer perceptron, and second a model based on frequent closed character sequences with a SVM. The BiLSTM system gave the best results by far. It ranked first on task 1, a binary theme classification task, and third on task 2, a four-class polarity classification task. This result is very encouraging as this method has very few priors, is completely endogenous, and does not require any specific preprocessing.
Complete list of metadata

Cited literature [13 references]  Display  Hide  Download
Contributor : Joseph Le Roux Connect in order to contact the contributor
Submitted on : Tuesday, January 22, 2019 - 10:21:26 AM
Last modification on : Friday, March 11, 2022 - 3:25:50 AM
Long-term archiving on: : Tuesday, April 23, 2019 - 1:51:09 PM


Files produced by the author(s)


  • HAL Id : hal-01988907, version 1


Davide Buscaldi, Joseph Le Roux, Gaël Lejeune. Modèles en Caractères pour la Détection de Polarité dans les Tweets. Atelier DEFT 2018, May 2018, Rennes, France. ⟨hal-01988907⟩



Record views


Files downloads