University Sétif 1 FERHAT ABBAS Faculty of Sciences
Détail de l'auteur
Auteur Amina Nourhane Ziad |
Documents disponibles écrits par cet auteur
Ajouter le résultat dans votre panier Affiner la recherche
Titre : Measuring and Mitigating Bias in Arabic Language Generation Type de document : texte imprimé Auteurs : Amina Nourhane Ziad, Auteur ; Chaima Mahdadi, Auteur ; Fouzi Harrag, Directeur de thèse Année de publication : 2022 Importance : 1 vol (53 f .) Format : 29cm Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Bias
Artificial intelligenceIndex. décimale : 004 Informatique Résumé :
Pre-training large language models on vast amounts of web-scraped text is a current trend
in natural language processing. While the resulting models are capable of generating convincing
text, they also reproduce harmful social biases. This thesis explores expressions
of bias in Arabic text generation.
Analyses will be performed on samples by the generative models AraGPT (GPT-2
[Radford et al., 2019] fine-tuned for Arabic). An Arabic classifier based on new transformer
model AraBERT ((BERT [Devlin et al., 2018] fine-tuned for Arabic) will be used
to captures the social Bias of an Arabic sentence or text. For the development of this classifier,
a dataset will be crowd-sourced, cleaned, and independently annotated. AraGPT
will be used to generate more biased descriptions from the standard prompts. Our bias
detection model will be based on the combination of the two transformers (AraGPTAraBERT)
models.
In addition to our quantitative evaluation study, we also conducted a qualitative study
to understand how our system would compare to others approaches where users try to
find bias in Arabic texts generated using AraGPt model. Taking into account that this
is the first study of its kind in the field of detection of bias in Arabic language generated,
we can say that our proposed model has achieved very encouraging results by reaching
an accuracy percentage of 81%.Côte titre : MAI/0608 En ligne : https://drive.google.com/file/d/1PjmG8MHtc_F1e25eGPG4p6LSRflzcXnv/view?usp=share [...] Format de la ressource électronique : Measuring and Mitigating Bias in Arabic Language Generation [texte imprimé] / Amina Nourhane Ziad, Auteur ; Chaima Mahdadi, Auteur ; Fouzi Harrag, Directeur de thèse . - 2022 . - 1 vol (53 f .) ; 29cm.
Langues : Français (fre)
Catégories : Thèses & Mémoires:Informatique Mots-clés : Bias
Artificial intelligenceIndex. décimale : 004 Informatique Résumé :
Pre-training large language models on vast amounts of web-scraped text is a current trend
in natural language processing. While the resulting models are capable of generating convincing
text, they also reproduce harmful social biases. This thesis explores expressions
of bias in Arabic text generation.
Analyses will be performed on samples by the generative models AraGPT (GPT-2
[Radford et al., 2019] fine-tuned for Arabic). An Arabic classifier based on new transformer
model AraBERT ((BERT [Devlin et al., 2018] fine-tuned for Arabic) will be used
to captures the social Bias of an Arabic sentence or text. For the development of this classifier,
a dataset will be crowd-sourced, cleaned, and independently annotated. AraGPT
will be used to generate more biased descriptions from the standard prompts. Our bias
detection model will be based on the combination of the two transformers (AraGPTAraBERT)
models.
In addition to our quantitative evaluation study, we also conducted a qualitative study
to understand how our system would compare to others approaches where users try to
find bias in Arabic texts generated using AraGPt model. Taking into account that this
is the first study of its kind in the field of detection of bias in Arabic language generated,
we can say that our proposed model has achieved very encouraging results by reaching
an accuracy percentage of 81%.Côte titre : MAI/0608 En ligne : https://drive.google.com/file/d/1PjmG8MHtc_F1e25eGPG4p6LSRflzcXnv/view?usp=share [...] Format de la ressource électronique : Exemplaires (1)
Code-barres Cote Support Localisation Section Disponibilité MAI/0608 MAI/0608 Mémoire Bibliothéque des sciences Anglais Disponible
Disponible