| 朱云琴,范静,蔡鸿宇.在线健康社区医生回复与生成式人工智能回复对比研究[J].数字图书馆论坛,2026,22(6):13~25 |
| 在线健康社区医生回复与生成式人工智能回复对比研究 |
| Research on Immersive Experience Service Models for Public Libraries Driven by AIGC Technology |
| 投稿时间：2026-05-29 |
| DOI：10.3772/j.issn.1673–2286.2026.06.002 |
| 中文关键词: 用户需求；在线健康社区；自我决定理论；GenAI；人工智能生成内容 |
| 英文关键词: User Needs; Online Health Communities; Self-Determination Theory; GenAI; Artificial Intelligence Generated Content |
| 基金项目:本研究得到教育部人文社会科学研究规划项目“基于智能化方法的在线医疗精准服务机理与策略研究”（编号：22YJA630018）、中央高校基本科研业务费专项资金资助项目“更有效的沟通：多模态信号与人格匹配如何塑造对话式AI的用户信任与参与”（编号：2025JX048）资助。 |
| 作者 | 单位 | | 朱云琴 | 北京外国语大学国际商学院 | | 范静 | 北京外国语大学国际商学院 | | 蔡鸿宇 | 南京大学信息管理学院 |
|
| 摘要点击次数: 9 |
| 全文下载次数: 12 |
| 中文摘要: |
| 生成式人工智能（Generative Artificial Intelligence，GenAI）拓展了健康咨询的信息供给渠道，但GenAI回复与在线健康社区医生回复的差异尚需系统考查。首先，本研究结合扎根理论和自我决定理论，对白白手拉手社区的用户提问进行编码，并利用大语言模型完成全部提问的需求分类。结果表明，用户需求可分为能力需求（医疗咨询）、关系需求（情感宣泄与社会融入咨询）与自主需求（生活方式与生育决策咨询）三类，其中能力需求占主导地位、关系需求与自主需求起补充作用。其次，针对各条提问使用GenAI生成回复，采用可读性指标配对检验与BERTopic主题建模，从信息可读性和主题维度对在线健康社区医生回复与GenAI回复进行对比。研究发现：在信息可读性维度，GenAI回复平均字数更多、表达方式更稳定，但可读性较差，而医生能够根据需求类型灵活调整回复表达，在面对复杂的能力需求时医生展现出更好的个性化适应能力；在主题维度，医生回复的主题分布多元、语义差异显著，主题词的语义核心突出，而GenAI回复主题高度集中、文档一致性更强，存在明显的语言同质化问题。上述差异表明，GenAI适合承担表达稳定、内容标准化的信息供给任务，而涉及个性化判断与情感支持的需求仍需医生回复。 |
| 英文摘要: |
| Generative artificial intelligence (GenAI) offers a new way to respond to users’ health-related questions, but differences between GenAI and physician responses in online health communities remain to be systematically examined. First, this study applies grounded theory and self-determination theory to code user questions posted in the Baibai Shoulashou community, and uses a large language model to classify all questions by need type. The results identify three categories of user needs: competence needs (medical consultation), relatedness needs (emotional expression and inquiries about social integration), and autonomy needs (inquiries about lifestyle and reproductive decisions). Competence needs predominate, while relatedness and autonomy needs play supplementary roles. Second, GenAI responses are generated for the questions. Paired tests of readability measures and BERTopic topic modeling are used to compare physician and GenAI responses in terms of readability and thematic characteristics. GenAI responses are longer on average and more consistent in their patterns of expression, but less readable. Physicians adjust their responses more flexibly across need types and demonstrate greater adaptability to individual circumstances when addressing complex competence needs. Thematic analysis shows that physician responses cover a more diverse range of topics, exhibit greater semantic differentiation, and have more salient core topic terms. GenAI responses are more concentrated by topic and show greater withintopic document consistency, but also exhibit marked linguistic homogeneity. These differences suggest that GenAI is better suited to providing information in a stable, standardized manner, whereas needs involving individualized judgment and emotional support continue to require physician responses. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
