你如何通过提问来获取信息：审核TikTok搜索中的乳腺癌错误信息
How You Ask Shapes What You Get: Auditing Breast-Cancer Misinformation in TikTok Search
中文总结 AI 辅助
研究TikTok搜索中用户查询框架对乳腺癌错误信息曝光的影响，通过受控实验发现查询框架与错误信息曝光紧密相关，替代医学查询错误信息返回率高，医学信息查询也有错误信息，强调审核查询驱动搜索系统的重要性。
中文摘要 AI 辅助
数百万人使用TikTok来寻求健康信息，但对于用户搜索查询如何影响健康错误信息的接触情况却知之甚少。以往算法审核主要关注推荐信息流，而本文研究TikTok的搜索系统，用户通过查询明确表达信息需求。通过对TikTok搜索进行受控的虚拟账号审核，将30个新账号分配到六个实验条件，涵盖三种信息寻求框架和两种乳腺癌背景。在9020次可用搜索结果曝光中发现，查询框架与错误信息曝光密切相关。替代医学查询在症状察觉和积极治疗背景下的癌症相关结果中，错误信息返回率分别为54.1%和53.5%，远高于医学信息查询。医学信息查询也有一定比例的可能错误信息。此外，替代医学查询的可能错误信息在整个排名搜索结果中都有出现。被标记为错误信息的视频更可能包含推广无支持治疗或反标准护理观点的评论。这些发现表明搜索查询框架在塑造TikTok上的错误信息曝光中起核心作用，并凸显了审核查询驱动搜索系统的重要性。
英文摘要
Millions of people use TikTok to seek health information, yet little is known about how users' search queries shape exposure to health misinformation. Whereas prior algorithm audits have focused primarily on recommendation feeds, we examine TikTok's search system, where users explicitly express their information needs through query formulation. We conduct a controlled sock-puppet audit of TikTok Search using 30 fresh accounts assigned to six experimental conditions spanning three information-seeking framings (Medical Information, Alternative Medicine, and Peer Narrative) and two breast-cancer contexts (Symptom Noticing and Active Treatment). Across 9,020 usable search-result exposures, annotated using a validated vision-language model pipeline, we find that query framing is strongly associated with misinformation exposure. Alternative Medicine queries returned misinformation in 54.1\% of cancer-relevant results within the Symptom Noticing context and 53.5\% within the Active Treatment context, 8.6 times and 7.6 times higher, respectively, than clinically framed Medical Information queries. Even Medical Information queries returned measurable levels of possible misinformation (6.3\%--7.1\%), suggesting that explicit medical intent alone does not eliminate exposure. Moreover, for Alternative Medicine queries, possible misinformation appeared throughout the ranked search results rather than only near the top, showing that exposure is not confined to the highest-ranked results. Videos labeled as misinformation were also substantially more likely to contain comments promoting unsupported treatments or anti-standard-care views. These findings demonstrate that search query framing plays a central role in shaping misinformation exposure on TikTok and highlight the importance of auditing query-driven search systems alongside recommendation algorithms.
