文章摘要
向菲,李俊豪.智能健康信息服务出错后回应配置对用户信任评价的影响研究[J].数字图书馆论坛,2026,22(7):11~20
智能健康信息服务出错后回应配置对用户信任评价的影响研究
How Post-Error Response Configurations Shape User Trust Evaluations in Intelligent Health Information Services
投稿时间:2026-06-12  
DOI:10.3772/j.issn.1673–2286.2026.07.002
中文关键词: 智能健康信息服务;出错后回应;错误后信任;用户信任评价;任务风险
英文关键词: Intelligent Health Information Services; Post-Error Response; Post-Error Trust; User Trust Evaluation; Task Risk
基金项目:本研究得到中央高校基本科研业务费项目“主动健康素养视域下老年人信息支持多重社会网络构建”(编号:2024WKYXQN006)资助。
作者单位
向菲 华中科技大学同济医学院医药卫生管理学院 
李俊豪 华中科技大学同济医学院医药卫生管理学院 
摘要点击次数: 4
全文下载次数: 1
中文摘要:
      生成式人工智能对话系统已应用于健康信息咨询等智能健康信息服务,但在系统出错被用户指出后,不同回应配置对用户信任的影响尚待研究。本研究开展两项2×3被试间实验(有效样本分别为541人和250人),研究不同错误情景与任务风险下生成式人工智能对话系统的回应配置对用户信任评分的影响。研究发现:在共同纠错的基础上,包含解释与改进承诺的综合回应信任评分高于“道歉+解释”和“道歉+改进承诺”两类单增量回应配置的平均得分;在系统异常情景下,改进承诺相较于解释的善意信任评分差,大于信息偏差情景下的该类优势;在高风险任务下,改进承诺相较于解释的总体信任评分差,大于低风险任务下的该类优势,但任务风险未显著放大综合回应的总体信任评分差。本研究揭示了回应配置的情景差异,为智能健康信息服务出错后的回应设计提供经验证据。
英文摘要:
      Generative AI dialogue systems are increasingly used to provide health information services. This study examines users’ post-error trust evaluations of different response configurations under different error scenarios and task-risk conditions after a system error has been identified. Holding error acknowledgement, core correction, and a professional-verification prompt constant, we compare apology plus explanation, apology plus improvement commitment, and a comprehensive response. Study 1 used a 2 (information bias vs. system anomaly)×3 between-subjects design (N=541); Study 2 used a 2 (low-vs. high-risk task)×3 design (N=250). Comprehensive responses produced significantly higher post-error trust ratings than the two single-increment configurations on average (0.36 and 0.48 points). Task risk did not significantly amplify this overall difference. Under system anomalies, the difference in benevolence trust ratings between commitment and explanation increased significantly under high-risk tasks, the difference in overall post-error trust ratings between commitment and explanation also increased significantly. This research highlights context-dependent differences in response configurations and provides empirical evidence for designing post-error responses in intelligent health information services.
查看全文   查看/发表评论  下载PDF阅读器
关闭