COVID-19 Persian Misinformation Detection on Instagram: A Comparative Analysis of Machine Learning and Deep Learning...

Area of Science:

• Computational Linguistics
• Social Media Analysis
• Public Health Informatics

Background:

• Social media misinformation, particularly on platforms like Instagram, poses risks to public trust and health.
• Existing research on misinformation detection is predominantly English-centric, leaving a gap in other languages.
• Detecting COVID-19 misinformation in Persian on Instagram is crucial for public health interventions.

Purpose of the Study:

• To establish a performance benchmark for detecting Persian-language misinformation on Instagram.
• To create and utilize a novel dataset of Persian comments for misinformation analysis.
• To evaluate various machine learning and deep learning models for this task.

Main Methods:

• A dataset of 27,000 Persian Instagram comments related to COVID-19 was curated and labeled.
• Five models were employed: XGBoost (TF-IDF), LSTM (GloVe), CNN, KNN, and BERT.
• Models underwent training and validation to assess performance and prevent overfitting, using confusion matrices for evaluation.

Main Results:

• The Long Short-Term Memory (LSTM) model demonstrated superior performance.
• LSTM achieved an accuracy of 0.97, precision of 0.91, recall of 0.85, and an F1-score of 0.85.
• Training and validation curves confirmed the LSTM model's effectiveness in mitigating overfitting.

Conclusions:

• This study presents the first high-performance baseline for deep learning-based misinformation detection on Persian social media content.
• The LSTM model offers a robust solution for identifying COVID-19 misinformation in Persian on Instagram.
• The findings highlight the importance of developing multilingual approaches to combat online misinformation.