Using deep learning to analyze the psychological effects of COVID-19

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Psychology

Background:

• Sentiment Analysis (SA) automates attitude classification using NLP.
• Deep learning shows promise in SA, but Arabic SA faces challenges due to language complexity and data scarcity.
• Anxiety as a psychological sentiment in Arabic research is under-explored.

Purpose of the Study:

• To address challenges in Arabic Sentiment Analysis (ASA).
• To develop a deep learning model for predicting anxiety levels during COVID-19 in Saudi Arabia.

Main Methods:

• Validated a psychological scale to measure anxiety.
• Created the Arabic Psychological Lexicon (AraPh) with 138 dialectical Arabic anxiety words.
• Annotated a corpus (Aranxiety) of 955 Arabic tweets using Bi-GRU with word embedding.

Main Results:

• The SA model achieved 88% accuracy, 89% precision, 88% recall, and 87% F1-score.
• 77% of tweets showed no anxiety, 17% mild anxiety, and 6% high anxiety.
• The model effectively analyzed sentiment and determined anxiety levels in tweets.

Conclusions:

• The proposed model can aid the Saudi Ministry of Health in enhancing psychological resilience.
• Findings offer insights into public anxiety levels during the COVID-19 pandemic in Saudi Arabia.