An Ensemble Deep Learning Model for Drug Abuse Detection in Sparse Twitter-Sphere.

Area of Science:

• Computational social science
• Machine learning applications
• Public health informatics

Background:

• Drug abuse monitoring increasingly relies on social media data, particularly Twitter.
• Classifying drug abuse-related tweets faces challenges due to data imbalance and slang, hindering accurate analysis.
• Existing machine learning models struggle with the rarity of abuse-related content in large, diverse datasets.

Purpose of the Study:

• To develop and evaluate an ensemble deep learning model for improved classification of drug abuse-related tweets.
• To address the challenge of imbalanced datasets in social media analysis for drug abuse studies.
• To compare the performance of the proposed deep learning model against traditional machine learning approaches.

Main Methods:

• Designed an ensemble deep learning model incorporating both word-level and character-level features.
• Utilized a Twitter dataset, simulating varying degrees of class imbalance (abuse vs. non-abuse tweets).
• Experimentally evaluated the model's efficacy in classifying drug abuse-related content under imbalanced conditions.

Main Results:

• The ensemble deep learning model demonstrated superior performance compared to ensembles of traditional machine learning models.
• Performance gains were particularly significant on datasets with high levels of class imbalance.
• The model effectively handles the classification of drug abuse tweets, including those with slang terms.

Conclusions:

• Ensemble deep learning offers a robust solution for analyzing imbalanced social media data in drug abuse research.
• The proposed model enhances the accuracy of identifying drug abuse-related activities on platforms like Twitter.
• This approach has significant implications for public health surveillance and intervention strategies.