关键词: 公共图书馆, 健康中国, 健康信息服务, DeepSeek, 服务质量评估, 数据挖掘

Abstract:

[Purpose/Significance] As part of the Healthy China strategy, public libraries are expected to serve as accessible hubs for health knowledge dissemination and community health literacy improvement. However, health information services are inherently high stakes: users' perceived credibility does not necessarily guarantee medical accuracy, and outdated or poorly contextualized information may lead to serious consequences. Evaluations in existing library settings still rely heavily on surveys and experience-based summaries. While these are useful for describing satisfaction, they are often insufficient for identifying process bottlenecks, explaining why specific dimensions perform poorly, or supporting the accountable governance of high-risk content. This study aims to develop an evidence-oriented and interpretable evaluation approach for public library health information services, and to demonstrate how DeepSeek can enhance the measurability and actionability of quality assessment beyond conventional questionnaire-and-interview approaches. [Method/Process] We selected 32 public libraries in Shanxi Province for our empirical study. A four-dimension, sixteen-indicator framework was established (Professionalism, Ease of Use, Timeliness, and Personalization) through literature consolidation, service-process decomposition, and expert consultation. Multi-source data were collected and aligned, including 1 842 valid questionnaires, approximately 2.4 million service log records, and about 5.6 million platform behavior records. DeepSeek was employed to transform unstructured service texts (online consultations, user feedback, activity reviews, and staff response summaries) into structured evidence: it consolidates theme–intent patterns, extracts reason-oriented points that explain dissatisfaction, and maps those textual cues to specific secondary indicators. In parallel, process features such as zero-result search rate, path length, response time, and repeated consultation rate were derived from logs to represent observable friction and efficiency. An intervention validation was conducted using a quasi-experimental pre-post design with matched experimental and control libraries over six months, applying evidence-driven improvements linked to low-performing indicators. [Results/Conclusions] The overall mean score of health information service quality was 3.43 out of 5, indicating a medium level across the sample. Structural analysis shows that professionalism and ease of use exert stronger effects on overall quality, implying that trustworthy content governance and accessible service pathways form the foundational layer of health information services. Personalization receives the lowest average score and the largest between-library variation, suggesting that differentiated entry design and feedback closure mechanisms are the main sources of divergence. Update frequency is positively associated with satisfaction, highlighting the practical importance of routine maintenance rather than campaign-style updates. In the intervention study, the experimental group achieved a 17.5% improvement in overall quality, significantly higher than the control group. The findings indicate that aligning perceived evaluations with behavioral evidence and text-based explanatory cues can improve both interpretability and implementability of quality assessment. Limitations include regional sampling and potential differences in logging completeness. Future work should expand cross-regional validation and incorporate outcome-oriented indicators that more accurately reflect problem resolution.

Key words: public library, healthy China, health information services, DeepSeek, service quality assessment, data mining