人工智能赋能中医体质辨识与方言语音交互的老年友好型健康服务平台：研究进展与构建思路

AI-Enabled Elderly-Friendly Health Service Platform for TCM Constitution Identification and Dialect Speech Interaction: Research Progress and Construction Framework

摘要: 人口老龄化与慢性病共病趋势使老年健康管理面临持续性、个体化和可及性要求，而以文字、普通话和触屏操作为主的数字健康平台难以覆盖不识字、方言使用及感官功能下降的老年群体。本文围绕老年健康服务需求、中医体质辨识与“治未病”理论、人工智能(AI)及方言语音交互技术，梳理相关研究进展，并提出老年友好型健康服务平台的构建思路。现有证据显示，机器学习可用于慢病风险预测与认知筛查，中医体质辨识为个体化生活方式干预提供了理论入口，方言语音交互则有助于降低健康信息采集的使用门槛。平台宜以方言语音为主要入口，以AI动态问卷和可穿戴设备为数据支撑，以体质与风险评估、专业复核、家庭协同和线下服务转介为输出，形成“采集–分析–干预–反馈”闭环。当前仍需重点解决方言老年语料不足、证候与体质量化困难、数据质量与算法公平、隐私保护及临床外部验证不足等问题。未来应通过多中心、前瞻性和真实世界研究，验证平台的可用性、准确性、健康结局与服务公平性。

Abstract: The concurrent trends of population aging and multimorbidity impose sustained, individualized, and accessible demands on elderly health management. However, digital health platforms that primarily rely on text, Mandarin, and touchscreen operations are inadequate for older adults who are illiterate, speak dialects, or experience age-related sensory decline. This paper reviews relevant research progress on elderly health service needs, Traditional Chinese Medicine (TCM) constitution identification and the theory of “preventive treatment of disease”, as well as artificial intelligence (AI) and dialect speech interaction technologies, and proposes a construction framework for an elderly-friendly health service platform. Existing evidence indicates that machine learning can be applied to chronic disease risk prediction and cognitive screening, TCM constitution identification provides a theoretical entry point for personalized lifestyle interventions, and dialect speech interaction helps lower the barrier to health information collection. The proposed platform should adopt dialect speech as the primary entry point, supported by AI-driven dynamic questionnaires and wearable devices for data acquisition, and deliver outputs including constitution and risk assessment, professional review, family coordination, and offline service referral, thereby forming a closed loop of “collection-analysis-intervention-feedback”. Key challenges remain to be addressed, including insufficient dialect speech corpora from elderly speakers, difficulties in quantifying TCM syndromes and constitutions, data quality and algorithmic fairness, privacy protection, and the lack of clinical external validation. Future efforts should focus on multicenter, prospective, and real-world studies to validate the platform’s usability, accuracy, health outcomes, and service equity.