Sentiment analysis in Arabicopinion polarity detection

  1. Rushdi Saleh, Mohammed
Dirigée par:
  1. María Teresa Martín Valdivia Directrice
  2. Luis Alfonso Ureña López Directeur

Université de défendre: Universidad de Jaén

Fecha de defensa: 07 octobre 2013

Jury:
  1. Ruslan Mitkov President
  2. José Manuel Perea Ortega Secrétaire
  3. José Antonio Troyano Jiménez Rapporteur
Département:
  1. INFORMÁTICA

Type: Thèses

Teseo: 363945 DIALNET lock_openRUJA editor

Résumé

Sentiment analysis is becoming increasingly important due the growing popularity of Web 2.0. This study focuses mainly on how to analyze opinions in Arabic language and predict their polarity. To achieve that, two corpora have been generated (OCA and EVOCA), OCA is an opinion corpus for Arabic movie reviews, while EVOCA is the translated version of OCA to English. Another corpus was created (SINAI-SA corpus) used with other corpora in order to predict sentiments in different domains. SINAI corpus was also used to study how to sort comments behave as textual information for the prediction of customer rates. Another question that was solved in this study is “How to treat with the neutral reviews”. Two main approaches have been investigated in this research, one based on semantic orientation and the other one based on machine learning algorithms like SVM or NB