Investigating the Key Success Factors of Chatbot-Based Positive Psychology Intervention with Retrieval- and Generative Pre-Trained Transformer (GPT)-Based Chatbots

Technologically assisted methods have been extensively utilized in positive psychology interventions (PPIs), which aim to elevate the happiness levels of the widespread and diverse general public, a population that traditional methodologies struggle to comprehend and impact. Nevertheless, the literature provides insufficient insights into the effectiveness of chatbot-based PPIs (Chat-PPIs). This study endeavors to fill this void by employing both retrieval-based and generative pre-trained transformer (GPT)-based chatbots and executing three randomized controlled trials involving 326 participants to investigate the hypothesized effectiveness mechanisms. The statistical analysis affirms the effectiveness of Chat-PPI. Moreover, our results indicate that personalized PPI recommendations, adaptive multi-round dialogues, and real-time feedback significantly augment the efficacy of Chat-PPI. Besides exploring and confirming the mechanisms behind Chat-PPI, this study also endorses the use of generative chatbots, which, although less controllable, provide more natural interaction that can boost the efficacy of Chat-PPI.