Identifying X (Formerly Twitter) Posts Relevant to Dementia and COVID-19: Machine Learning Approach

Area of Science:

• Computational linguistics
• Artificial intelligence in healthcare
• Social media analytics

Background:

• Dementia patients are a vulnerable population during pandemics.
• Social media platforms like X (formerly Twitter) are key information sources for COVID-19 updates.
• Identifying dementia-related posts is crucial for supporting patients and caregivers but challenging due to data volume.

Purpose of the Study:

• To automate the identification of social media posts relevant to dementia and COVID-19.
• To leverage natural language processing and machine learning for this task.

Main Methods:

• Utilized natural language processing and machine learning algorithms.
• Employed manually annotated posts for training and validation.
• Assessed algorithm performance on over 100,000 posts across three datasets.

Main Results:

• Transfer learning algorithms significantly outperformed traditional machine learning methods.
• The ALBERT (A Lite Bidirectional Encoder Representations from Transformers) model achieved 82.92% accuracy and 83.53% AUC.
• ALBERT demonstrated superior performance in classifying relevant posts.

Conclusions:

• Transfer learning models, such as ALBERT, are highly effective for topic-specific social media post identification.
• These algorithms excel even with limited or adjacent data, outperforming other machine learning approaches.
• Automated analysis reduces manual coding workload, aiding researchers and policymakers in supporting vulnerable groups.