Investigating the Relationship Between Physical Activity and Tailored Behavior Change Messaging: Connecting Contextual Bandit with Large Language Models (2506.07275v2)

Abstract: Machine learning approaches, such as contextual multi-armed bandit (cMAB) algorithms, offer a promising strategy to reduce sedentary behavior by delivering personalized interventions to encourage physical activity. However, cMAB algorithms typically require large participant samples to learn effectively and may overlook key psychological factors that are not explicitly encoded in the model. In this study, we propose a hybrid approach that combines cMAB for selecting intervention types with LLMs to personalize message content. We evaluate four intervention types: behavioral self-monitoring, gain-framed, loss-framed, and social comparison, each delivered as a motivational message aimed at increasing motivation for physical activity and daily step count. Message content is further personalized using dynamic contextual factors including daily fluctuations in self-efficacy, social influence, and regulatory focus. Over a seven-day trial, participants receive daily messages assigned by one of four models: cMAB alone, LLM alone, combined cMAB with LLM personalization (cMABxLLM), or equal randomization (RCT). Outcomes include daily step count and message acceptance, assessed via ecological momentary assessments (EMAs). We apply a causal inference framework to evaluate the effects of each model. Our findings offer new insights into the complementary roles of LLM-based personalization and cMAB adaptation in promoting physical activity through personalized behavioral messaging.

• The paper shows how combining cMAB algorithms with LLMs improves the personalization of physical activity interventions.
• The study employs randomized trials and causal inference to evaluate user acceptance and changes in daily step counts.
• The findings suggest that integrating AI techniques can enhance adaptive health messaging and inform future digital health research.

Investigating the Relationship Between Physical Activity and Tailored Behavior Change Messaging

The research presented in this paper aims to explore the potential of integrating contextual multi-armed bandits (cMABs) and LLMs to enhance personalized interventions for promoting physical activity. The intersection of these two domains promises to address the challenges inherent in tailoring health behavior change interventions, particularly in capturing the dynamic psychological contexts that influence individual receptiveness to motivational messaging.

The paper employs cMAB algorithms to select appropriate intervention types based on contextual factors such as self-efficacy, social influence, and regulatory focus. The intervention types considered include behavioral self-monitoring, gain-framed, loss-framed, and social comparison messages. LLMs are then used to tailor the language of these interventions, ensuring that the messages are linguistically aligned with the recipient’s psychological state.

The experimental framework involves four models: (1) a Randomized Controlled Trial (RCT) as a baseline, (2) cMAB alone, (3) LLM alone, and (4) a combined cMABxLLM approach. In a seven-day trial, participants receive daily motivational messages through one of these models, and outcomes are measured via daily step count and message acceptance using ecological momentary assessments (EMAs). By employing a causal inference framework, this paper assesses the effects of different intervention models on physical activity and message acceptance, providing novel insights into personalized digital health interventions.

The paper is anchored in addressing two primary research questions: the efficacy of personalization via cMAB and LLM in terms of user acceptance and its impact on motivation for physical activity. Initial findings suggest that the combination of cMAB with LLM personalization (cMABxLLM) holds promise for crafting interventions that are both adaptive to shifting motivational states and effective in sustaining engagement. Though the current paper is a small-scale pilot with a limited sample size, it sets the groundwork for future research targeting larger populations.

In terms of practical implications, the research underlines the potential for integrating ML techniques to refine the delivery of digital health interventions. The hybrid approach proposed, which leverages the contextual sensitivity of cMABs with the flexible linguistic capabilities of LLMs, could serve as a foundation for designing more nuanced health feedback systems. On a theoretical level, the paper contributes to the understanding of how various psychological and situational parameters can be strategically harnessed to enhance intervention efficacy.

Looking forward, one can anticipate the role of AI in personalized health applications expanding significantly. Future studies should focus on overcoming the limitations noted, such as small sample sizes and short intervention durations, and on refining the adaptive algorithms used to ensure reliable personalization across diverse contexts and populations. Furthermore, exploring the integration of more comprehensive data sets, potentially leveraging synthetic data to augment learning, could mitigate the data demand challenge in the deployment of cMAB algorithms.

In summary, this paper provides valuable insights into the evolving methodology for deploying tailored behavioral interventions and lays the groundwork for future research that can leverage AI to address public health challenges related to physical inactivity and sedentary lifestyles.