Des représentations continues de mots pour l'analyse d'opinions en arabe: une étude qualitative

Abstract : Word embeddings for Arabic sentiment analysis : a qualitative study In this paper, we are interested in Arabic sentiment analysis task. Recently, the use of deep learning improves many automatic systems in a wide variety of fields (image analysis, speech recognition, machine translation,. . .), among others English sentiment analysis. Thus, we study the performance of two architectures (CNN and LSTM) in our specific framework. In addition, we investigated the use of several types of word embeddings publically available for Arabic, that achieve good results. Finally, the analysis of the errors of our system and the relevance of the different embeddings was also proposed. These analysis lead to several interesting perspectives : building expert resources (lexicon) and relevant task-specific embeddings.
Document type :
Conference papers
Complete list of metadatas

Cited literature [20 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01757776
Contributor : Amira Barhoumi <>
Submitted on : Wednesday, March 6, 2019 - 9:50:03 AM
Last modification on : Friday, April 26, 2019 - 1:35:02 PM
Long-term archiving on : Friday, June 7, 2019 - 5:51:11 PM

File

Des_repr_sentations_continues_...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01757776, version 2

Collections

Citation

Amira Barhoumi, Nathalie Camelin, Yannick Estève. Des représentations continues de mots pour l'analyse d'opinions en arabe: une étude qualitative. 25e conférence sur le Traitement Automatique des Langues Naturelles (TALN 2018), May 2018, Rennes, France. ⟨hal-01757776v2⟩

Share

Metrics

Record views

41

Files downloads

62