User Modeling and Planning for Improving Self-efficacy and Goal Adherence in mHealth

Frontiers in Public Health

卷:4

Peter Pirolli¹ 

[1] Palo Alto Research Center;

关键词: artificial intelligence;    ACT-R;    Predictive Modeling;    mHealth;    Cognitive models;   

DOI  :  10.3389/conf.FPUBH.2016.01.00107

来源: DOAJ

【 摘 要 】

BACKGROUNDmHealth applications provide great opportunities for projecting behavior-change methods into everyday life at large economies of scale. We have developed a smartphone-based system that provides support for health behavior change in diet, nutrition, and stress reduction. A key component of the system is an artificially intelligent (AI) coaching agent that personalizes the selection of behavioral goals (e.g., walking for 30 minutes). The aim of this personalization is to maximize daily goal adherence as well as long-run physical and psychosocial gains. We describe two modules that are central to this personalized coaching: (1) an AI planner that dynamically adjusts the schedule of future goals in reaction to a user’s compliance with the schedule and (2) a predictive cognitive model of self-efficacy based on neurocomputational theory that continuously changes in reaction to individual achievements or failures.DESCRIPTIONThe planner is designed to encourage users to increase their aerobic activity to American Heart Association recommended guidelines (low: 30 minutes of moderate activity five times a week; high: 30 minutes of vigorous activity five times a week). The planner is based on F.I.T.T. (Frequency, Intensity, Time/Duration, Type) principles of exercise prescription. It operates over aerobic (Type) activities with varying intensity (i, measured in MET levels—Metabolic Equivalent of Tasks—a physiological measure expressing the energy cost of physical activities) including walking at moderate pace, walking on inclines, brisk and moderate interval walking, and walking with exercises.These activities can be scheduled for increasing duration i.e., time (d minutes) and frequency (f times a week), which are computed based on a user’s compliance with the past schedule. The planner plans at two temporal levels: weekly and daily. The weekly planner includes three components:-An assessor that interacts with a user to measure their pre-intervention ability (in i, d, f), their projected ability for the first week, and generates an initial plan to a relevant AHA goal. The plan is expressed as a set of weekly progressive goals (in i, d, f). -A heuristics-based evaluator that, based on a user’s compliance history, judges if the current plan is appropriate for the user in making progress toward their goal. If not, it suggests a revision to the plan which could be a progression (make the plan slightly harder), a regression (make the plan slightly easier), or a hold (repeat this week’s goal next week). -A reviser that changes the weekly plan in accordance with the evaluator’s recommendations. Revisions are made if the user has been successful at this week’s goal so that the next week includes an increase in i, d, or f. If the user is failing, the reviser may change the next week’s goal easier by backing off in one of these dimensions. Once a weekly plan is computed, the user is asked to confirm the goal if they feel they can pursue it. Otherwise, they are given an option to continue the last week’s goals. On confirmation, the daily planner distributes the activities on specific days. The daily planner acts to provide enough opportunity to users to achieve their weekly goals. As users report compliance or non-compliance, the daily planner redistributes activities in the remaining days. As users report on their daily activities, they are asked a few questions about the activity (e.g., How difficult was an activity? How likely they think they are to do an activity in future?). The responses to these questions are incorporated in the evaluator’s judgments and influence how future weekly goals are generated. The planner incorporates a user model to make predictions. The user model includes components that capture and predict both physical and psychological states of individuals. Physical abilities of users are based on observed achievement of activity goals that have known MET levels or other known empirical estimates difficulty levels. The psychological user model focuses on levels of self-efficacy that change dynamically in response to day-to-day achievements of personalized goals.This user model of self-efficacy is based on a neurocognitive model that refines the "macro-theory" of self-efficacy, as developed in Social Cognitive Theory, into a micro-theory implemented in the ACT-R cognitive simulation system. Predictions of levels of self-efficacy emerge from neurocognitive simulations of the interaction of individuals’ behavioral goals, memories of past experiences, and behavioral performance. Our ACT-R models of self-efficacy in multi-week mHealth exercise programs shows excellent predictive fits to individual daily exercise achievements. The user model can be consulted to predict the likelihood that any particular exercise can be performed by a particular user, and to predict the impact of success of doing an exercise (or failure) on future levels of self-effiacy.CONCLUSIONSOur user modeling efforts have been primarily focused on refining predictions of psychosocial changes that underlie the achievement of behavior-change goals. We have chosen to use the ACT-R neurocognitive theory and simulation environment as a way of driving the user modeling efforts because we believe that: (a) neurocognitive architectures provide a unified account of how the modules of the mind function together to produce coherent behavior, and provide an integrative explanation of data produced across specialized domains of psychology;(b) longer-term behavior change occurring over days, weeks, or months, can be decomposed to learning events occupying much briefer units of time; and (c) models in neurocognitive architectures provide a basis for bridging the events at the small scale to the dynamics of behavior change occurring at the large scale.Our planner is an adaptive, computational approach to the problem of interactive planning for health-behavior change. The approach ensures that the generated plan is tailored to each individual user and is reactive to their experience with it. This can lead to better compliance and long-term goal achievement.

【 授权许可】

Unknown