Documents disponibles écrits par cet auteur

Machine Learning and Deep Learning for Moroccan Dialect Sentiment Analysis / El Mahdi Mercha

Public ISBD Titre : Machine Learning and Deep Learning for Moroccan Dialect Sentiment Analysis Type de document : projet fin études Auteurs : El Mahdi Mercha, Auteur Langues : Français (fre) Catégories : BIG DATA Index. décimale : mast 256/19 Résumé : Sentiment analysis or opinion mining is a research area of natural language processing, that focus on the study of people's opinions, sentiments, emotions and attitudes about several entities such as products, services, issues, events and their attributes. In recent years, there exist a lot of research in the eld of sentiment analysis through exploiting messages shared in the social networks and written in English or in French etc. However, few studies have been carried out for sentiment analysis based on Arabic dialectal languages especially the Moroccan one. With frequent transcriptions switching between MSA, Latin and alphanumeric transcription by its writers, the processing of this dialect becomes even more complicated. YouTube is commonly used in Arab countries and specically in Morocco, we are interested, in this research, in the sentiment analysis of Moroccan user's comments on several YouTube videos. In this master thesis, we will explore the eld of Moroccan dialect sentiment analysis. First, we will describe the steps pursued to construct the Moroccan Dialect Sentiment Analysis Corpus (MDSAC). Second, we will carry out several sets of experiments to explore the impact of the dierent (i) settings for text representation, and (ii) Machine learning algorithms used. Two folds of studies have been conducted. The performance of classical M.L algorithms, namely Support Vector Machines (SVM), Naïve Bayes (NB) and Logistic regression (LR), in parralel with dierent settings for text representation, such as stemming type, indexing and weighting schemes, has been investigated. The second fold concerns the usage of deep learning for both text representation and classication. Two word embedding models were built, one using Continuous Bag of Words (CBoW), and the other using Skip-gram. The dierent architectures of deep learning ( CNN and LSTM) using the two dierent learned word representation were compared. For both folds, a large set of experiments was conducted to tune the dierent hyperparameters. Machine Learning and Deep Learning for Moroccan Dialect Sentiment Analysis [projet fin études] / El Mahdi Mercha, Auteur . - [s.d.]. Langues : Français (fre) Catégories : BIG DATA Index. décimale : mast 256/19 Résumé : Sentiment analysis or opinion mining is a research area of natural language processing, that focus on the study of people's opinions, sentiments, emotions and attitudes about several entities such as products, services, issues, events and their attributes. In recent years, there exist a lot of research in the eld of sentiment analysis through exploiting messages shared in the social networks and written in English or in French etc. However, few studies have been carried out for sentiment analysis based on Arabic dialectal languages especially the Moroccan one. With frequent transcriptions switching between MSA, Latin and alphanumeric transcription by its writers, the processing of this dialect becomes even more complicated. YouTube is commonly used in Arab countries and specically in Morocco, we are interested, in this research, in the sentiment analysis of Moroccan user's comments on several YouTube videos. In this master thesis, we will explore the eld of Moroccan dialect sentiment analysis. First, we will describe the steps pursued to construct the Moroccan Dialect Sentiment Analysis Corpus (MDSAC). Second, we will carry out several sets of experiments to explore the impact of the dierent (i) settings for text representation, and (ii) Machine learning algorithms used. Two folds of studies have been conducted. The performance of classical M.L algorithms, namely Support Vector Machines (SVM), Naïve Bayes (NB) and Logistic regression (LR), in parralel with dierent settings for text representation, such as stemming type, indexing and weighting schemes, has been investigated. The second fold concerns the usage of deep learning for both text representation and classication. Two word embedding models were built, one using Continuous Bag of Words (CBoW), and the other using Skip-gram. The dierent architectures of deep learning ( CNN and LSTM) using the two dierent learned word representation were compared. For both folds, a large set of experiments was conducted to tune the dierent hyperparameters.

Réservation

Réserver ce document

Exemplaires

Code barre	Cote	Support	Localisation	Section	Disponibilité
mast 256/19	mast 256/19 ELM	Texte imprimé	Unité des masters	Mast/19	Disponible