AI与行为科学双驱动的体重管理模型研究

Research on a Dual-Driven Weight Management Model Based on Artificial Intelligence and Behavioral Science

摘要: 当前国民肥胖及慢性病对国家公共健康的威胁日益严峻，传统体重管理模式因缺乏个性化指导，难以维持用户的长期参与。为此，本研究构建了一种人工智能和行为科学双驱动的个性化体重管理模型。研究通过设计融合计划行为理论与自我决定理论的多模态研究问卷，采集了包括饮食图像、运动文本及心理状态等多元化数据，并依托ResNet-50与BERT模型分别提取了图像与文本特征。在此基础上，采用样本与特征对齐机制及监督对比学习方法，强化了模型对健康与非健康行为的判别能力，最终利用多层感知机制生成专属化的运动与饮食建议。实证阶段共回收712份有效问卷，测试结果显示，模型在健康行为分类任务中的准确率达85.4%，行为依从性预测的平均绝对误差为0.23，用户的总体接受度为89.2%。这一研究结果证实，该模型能够有效融合多模态感知的信息并实现精准的行为干预，为打造具有持续依从性的智能健康管理系统提供了有力的理论支撑与实践路径。

Abstract: At present, the threats posed by national obesity and chronic diseases to public health are becoming increasingly severe. Traditional weight management models are difficult to maintain users’ long-term participation due to the lack of personalized guidance. To address this problem, this study constructs a personalized weight management model dual-driven by artificial intelligence and behavioral science. A multimodal research questionnaire integrating the Theory of Planned Behavior and Self-Determination Theory was designed to collect diversified data, including diet images, exercise texts and psychological states. The ResNet-50 and BERT models were used to extract image and text features respectively. On this basis, the sample and feature alignment mechanism and supervised contrastive learning method were adopted to enhance the model’s ability to distinguish between healthy and unhealthy behaviors. Finally, a multi-layer perceptron was utilized to generate personalized exercise and diet recommendations. In the empirical phase, a total of 712 valid questionnaires were collected. The test results show that the model achieves an accuracy of 85.4% in the healthy behavior classification task, the mean absolute error of behavior compliance prediction is 0.23, and the overall user acceptance rate reaches 89.2%. The research results confirm that the model can effectively integrate multi-modal perception information and realize precise behavioral intervention, which provides a strong theoretical support and practical path for building an intelligent health management system with sustained compliance.