Building an emotion regulation recommender algorithm for socially anxious individuals using contextual bandits

Area of Science:

• Psychology
• Computer Science
• Digital Health

Background:

• Poor emotion regulation (ER) is linked to mental health conditions like social anxiety disorder.
• Developing scalable interventions for ER is crucial for widespread mental health support.

Purpose of the Study:

• To create a novel contextual recommender algorithm for ER strategies.
• To assess the algorithm's effectiveness in improving ER for individuals with social anxiety.

Main Methods:

• 114 participants with social anxiety used a mobile app to report emotions, ER strategy use, and context over five weeks.
• Compared a contextual recommender algorithm against random, fixed (cognitive reappraisal), and observed ER strategies.
• Utilized contextual bandits to predict strategy effectiveness based on context.

Main Results:

• The contextual algorithm was not superior when ER strategies were broadly categorized.
• However, when considering the ten most frequently used strategies individually, the contextual algorithm significantly outperformed all other methods.
• Contextual tailoring of ER strategies proved more effective than indiscriminate use of cognitive reappraisal.

Conclusions:

• Categorizing ER strategies can mask context-specific effectiveness.
• Personalized, context-aware ER strategy recommendations are more effective than generalized approaches.
• Future research will evaluate the algorithm's real-world efficacy as a just-in-time intervention.