應用聚類分析於佛教醫學文獻主題探索之研究

佛教相當重視醫學知識與技術上的學習，從五明的「醫方明」即可看出對其的重視。佛教醫學的目的是治癒眾生由貪瞋痴三毒引發在身體與心靈上的疾病，使眾生最終能趨向、成就佛道。由此讓我們知道人的「病苦」是佛教與醫學都關注的議題；從而將佛教與醫學緊密的連結在一起。佛教醫學在《佛教醫學佛教圖書分類法》當中，被歸於應用學科之下。可見文獻之豐富、此主題之重要；在佛教典籍中，與佛教醫學相關的文獻著實不少，但往往零散而不具系統，在法鼓文理學院所製作的法的療癒資料庫中，內容共收錄305篇相關於佛教醫學的主題文章，但其文章分類架構，因仰賴人工的方式來進行，致使分類結果標準不明確，而導致分類結果不夠全面性。為解決以上問題，我們嘗試藉由資訊技術常用於資料分析的相關手法，也就是「文本探勘」（Text Mining）當中的「聚類分析」（Cluster Analysis）技術，針對法的療癒資料庫中所篩選出《大正新脩大藏經》律部的90卷經文內容為研究對象，進行「佛教醫學」的主題文件分群（text clustering），分群之後，藉由分析每個群組的共通關鍵用詞，以人工或自動的方式針對各個分群進行命名，嘗試建立相關文件的主題分類結構。本研究成功將結果分為十群，並於論文內詳加討論各群文章大意與群聚特性。雖無法明確給予各群一個獨立的命名，但在藉由資訊技術的輔助以量化統計方法，來建立有主題且結構化的佛教醫學律典文獻呈現的角度上，完成了初步嘗試與基礎貢獻。

關鍵字

佛教醫學；聚類分析；文本探勘；主題萃取；結構化文件

並列摘要

Buddhism attach great importance to the learning of medical knowledge and skills; the importance of it could be observed in “five sciences” (pañcavidyā) of ancient India, in which one of them is the “science of medicine” (cikitsā vidyā). The purpose of ‘medical learning’ in Buddhism, is to cure illness triggered by greed, aversion, and delusion (three poison), inflicted on body and mind of all living beings. Eventually, all ‘cured’ sentient beings will be guided and walk on the path to Buddhahood. Thus, Buddhism and Medical highly concern ‘suffering from illness’ of sentient beings. Though there are numerous texts related to medical in Buddhist scriptures, it is often scattered with no clear organization. The Dharma-Healing Database created by Dharma Drum Institute of Liberal Arts included 305 topic articles related to Buddhist medicine; however the categorization of the topic article relies on manual work, the standard of categorization became unclear, and the results of the categorization were not comprehensive. To solve the problem stated above, we attempted to apply hierarchical clustering in The Dharma-Healing Database and filtered out in the 90 volumes of scriptures from the Vinaya section of Taishō Tripiṭaka as the research object. Text clustering was applied to the Buddhist Medicine topic articles to create clusters. Through analyzing common keywords of each cluster, each cluster was given labels either manually or automatically. We then attempted to create a categorization structure for the topic articles. The research successfully established 10 clusters. Through the quantitative statistical methods, a basic structural categorization Buddhist Medicine text in Vinaya Pitaka was completed.