檢視台灣學者在醫療資訊領域學術研究傑出性排名：不同集群分析演算法比較

Examining the rankings of excellence in academic researches among Taiwanese scholars: A comparison of various clustering algorithms

醫療資訊促進醫學領域的研究與進步，探討台灣學術在該領域中的10年論文成果，利用傳統書目計量學的方法，僅係陳列各論文主體（如機構、期刊）的發表量與被引用量做排名與比較，其缺點是耗用過多的圖表與篇幅（如某論文列出42張圖表），卻未發掘研究問題與驗證其研究假設，致使其論文不具聚焦與可讀性。本研究收集2015至2024年間，台灣作者發表於醫療資訊SCI期刊的論文，利用簡要圖表，（1）闡述10項論文主體前10項目的成就排名，（2）驗證演算法皆符合集群原則且具解釋力的假設，（3）示範共字分析。集群原則包括：（1）每群集至少需包含3成員；（2）群集領導者（係論文量最高者）至少需有兩名追隨者（不包括排名最高者）；（3）每追國者僅連結一名領導者，係以關聯值最大者為依據。論文作者機構合作分析，建立其集群關聯，分析11常用集群演算法的分類結果，比較演算法符合集群規則與解釋力。結果顯示：（1）台灣作者於10年間在34醫療資訊SCI期刊發表868篇論文，h指數為52，（2）台北醫學大學發表論文量最多，其次為台灣大學（n=118及92，h指數為25及21），（3）多數演算法符合其3項集群原則，主從關係演算法（FLCA）富解釋力。本研究運用書目計量學的方法，列出近10年台灣作者在醫療資訊領域中10項論文主體前10項目的成就排名，且發現僅FLCA符合集群原則與具解釋力。未來研究可利用簡要圖表呈現論文主體的前10排名，與驗證其研究的假設，精簡書目計量學的論文內涵，改善傳統耗用過多的圖表與篇幅的缺失，提供其參考的示範。

Bibliometric studies in medical informatics often rely on long lists and many figures that rank outputs (e.g., institutions, journals) but seldom test hypotheses about structure, reducing focus and readability. To provide concise, hypothesis-driven insights into Taiwan-affiliated medical informatics publishing (2015-2024) by (1) reporting top-10 rankings across 10 bibliometric entities with minimal visuals, (2) evaluating whether common clustering algorithms satisfy explicit cluster-formation rules and yield interpretable structures, and (3) demonstrating co-word analysis. We compiled Taiwan-affiliated papers in SCI-indexed medical informatics journals (2015-2024), constructed an institutional collaboration network, and ran 11 clustering algorithms. We tested compliance with three rules: (i) each cluster contains 2:3 members; (ii) the cluster leader (highest output) has 2:2 followers; (iii) each follower links to exactly one leader via its maximum association. We compared algorithms on rule compliance and interpretability, highlighting the Follower-Leader Clustering Algorithm (FLCA). Taiwan-affiliated authors published 868 articles across 34 journals (overall h-index = 52). Taipei Medical University and National Taiwan University led output (118 and 92 papers; h-indices 25 and 21, respectively). Most algorithms satisfied some rules; FLCA uniquely satisfied all three while providing clear leader-follower structures and the most interpretable groupings. A compact co-word analysis further illustrated thematic organization without excessive Figures. Pairing concise top-10 reporting with rule-validated clustering produces focused, readable bibliometric assessments. In our 10-year Taiwan case, FLCA best aligned with the predefined cluster principles and delivered superior interpretability. This streamlined approach can reduce Figure bloat, sharpen hypothesis testing, and be generalized to other domains of bibliometric evaluation.