How digital therapeutic alliances influence the perceived helpfulness of online mental health Q&A: An explainable machine learning approach

Objective This study investigates the role of digital therapeutic alliance (DTA) in predicting and explaining the perceived helpfulness of responses on online mental health Q&A platforms. Methods This study constructs a large dataset of 19,682 Q&A interactions from online mental health Q&A platforms, employs natural language processing, explainable machine learning, and causal inference methods to identify and understand the factors, particularly DTA, that influence the perceived helpfulness of human counselors’ responses to mental health questions. Results The machine learning-based model for predicting perceived helpfulness demonstrated strong performance, achieving an root mean square error of 0.8234 and a mean absolute percentage error of 22.7288%. The explanatory analysis revealed that peripheral path-related language cues, such as counselor engagement (e.g., word count and response time), had the highest predictive power. Additionally, central path-related language cues, such as those linked to the DTA—specifically emotional bonds and therapeutic tasks—significantly influenced perceived helpfulness and were positively impacted by counselor engagement. Conclusion This study integrates DTA and elaboration likelihood model theories to propose a computational framework for understanding and predicting the perceived helpfulness of responses in online mental health Q&A platforms. Findings offer theoretical insights into the mechanisms of perceived helpfulness and practical guidance for optimizing platform design, training counselors, and improving user satisfaction through targeted language strategies.