Automatic detection of problem-gambling signs from online texts using large language models

Area of Science:

• Computational linguistics
• Psychology
• Public Health

Background:

• Problem gambling is a significant public health issue with severe psychological and economic consequences.
• Online gambling communities serve as platforms for information exchange and may reflect problem gambling behaviors.
• Individuals with higher levels of problem gambling are more active in these online communities.

Purpose of the Study:

• To fine-tune a Bidirectional Encoder Representations from Transformers (BERT) model for predicting problem gambling indicators from online forum posts.
• To assess the efficacy of a BERT-based model in detecting problem gambling signatures within user-generated content.
• To explore the potential of computational approaches for monitoring trends in online problem gambling.

Main Methods:

• Data was collected from a German online gambling discussion board.
• A Bidirectional Encoder Representations from Transformers (BERT) model was fine-tuned using manually annotated data.
• Training data incorporated diagnostic criteria and cognitive distortions associated with gambling.
• Cross-validation was employed to evaluate model performance.

Main Results:

• The BERT-based model achieved a precision of 0.95 and an F1 score of 0.71.
• Satisfactory classification performance was demonstrated through high-quality, manually annotated training data.
• The study confirmed the reliability of BERT models for analyzing small datasets in detecting problem gambling indicators.
• The model successfully identified signatures of problem gambling in online communication data.

Conclusions:

• Bidirectional Encoder Representations from Transformers (BERT) models can be effectively utilized to detect problem gambling from online forum posts.
• High-quality manual annotation based on diagnostic criteria is crucial for training accurate models.
• Computational methods, like the developed BERT model, offer a promising avenue for early detection and monitoring of problem gambling prevalence online.