ارسل ملاحظاتك

ارسل ملاحظاتك لنا









Using Machine Learning to Analyze Emotions in Arabic and Dialectical Texts

المصدر: المجلة العلمية لجامعة الملك فيصل - العلوم الأساسية والتطبيقية
الناشر: جامعة الملك فيصل
المؤلف الرئيسي: Hamed, Dina Abdelnaser (Author)
مؤلفين آخرين: Ben Bella, Said Tawfik (Co-Author) , Makhlouf, Mohamed Abdullah (Co-Author)
المجلد/العدد: مج25, ع2
محكمة: نعم
الدولة: السعودية
التاريخ الميلادي: 2024
التاريخ الهجري: 1445
الصفحات: 22 - 30
ISSN: 1658-0311
رقم MD: 1524328
نوع المحتوى: بحوث ومقالات
اللغة: الإنجليزية
قواعد المعلومات: science
مواضيع:
كلمات المؤلف المفتاحية:
Text Classification | Fine-Tuning | Voting Technique | Naive Bayes | Augmentation | Transformers
رابط المحتوى:
صورة الغلاف QR قانون
حفظ في:
المستخلص: Social media is an imperative necessity in contemporary life. People can easily express their emotions and share moments on social media by writing a few words. Organizations approach Twitter as a rich data source that may be used to study emotions, but while many efforts have focused on sentiment analysis from text, emotion classification has received less attention. Emotion analysis usually provides a more in-depth assessment of the author's feelings, and in this research, we propose a dialectal Arabic text emotion classification architecture that accurately classifies the expressions into four emotions (anger, joy, fear, and sadness). Considering the improvements in natural language processing (NLP), we investigated the Bidirectional encoder representations from transformers (BERT) model. We implemented our proposed ensemble model via a majority voting technique that merges the best three versions of the pre-trained BERT models that are considered state-of-the-art in the classification field. We compared the results of our model with eight other machine learning classifiers and ten versions of the BERT model. The proposed ensemble approach accomplished around 84%, however the highest accuracy of the other investigated models was 76%. The presented experiments were examined on the Arabic tweets’ dataset for the EI-OC task provided by SemiEval, which contains 5600 tweets.

ISSN: 1658-0311

عناصر مشابهة