Subject analysis

本文運用臺灣大學「臺灣文獻文物典藏數位化計畫」建置的資料庫，就臺灣古拓本與《淡新檔案》中的特定案例，進行相互參照的研究。臺灣大學圖書館典藏豐富，其中的古碑拓本涵蓋行政、建設、宗教、民俗等主題，資料型式為單件；《淡新檔案》為清代北臺灣的淡水廳、臺北府及新竹縣的地方行政及司法文書，資料型式為多份單件構成一個案件。該批此二種不同來源之史料產生時空環境相近，二者為研究上的相輔相成素材。然因拓本資料原件尺寸偏大，不便利用；而《淡新檔案》數量龐大，僅利用複製紙本或微縮資料，亦不易找出特定事件之相關案件。臺灣大學圖書館將此二種資料典藏數位化並建置資料庫之後，透過於資料庫中的搜尋，很容易可以比對出各資料群中關於相同事件之史料，而促成參照利用不同的史料進行研究之可能。本文擇一臺灣古碑拓本的「嚴禁自盡圖賴碑記」之例，透過整理《淡新檔案》中的相關公文案卷，重建當時臺灣地方政府的行政程序，並探討「自盡圖賴」事件背後所反映的社會意義。作者進一步為臺灣古碑拓本與《淡新檔案》等二種清季臺灣文獻，勾勒參照性比對研究的其他可能，將臺灣大學典藏的古碑拓本，逐一對照《淡新檔案》可能相關的類目，載於文末附錄，以供研究者參考。

This research was conducted by using the database established under the Digital Archives Project on Taiwan Historical Collections of National Taiwan University. The authors took a special case "Forbidden of Suicide in Fraud of Others" which appeared in both the Taiwan University's rubbings collection and the Tan-Hsin Archive as an example, and demonstrated how cross-referencing use of these two collections could help to uncover some historical problems of the Taiwan society and the solutions from the government. The monumental stone rubbings, mostly single autonomous items, depict various aspects such as politics, social development, religion, customs and culture of the early Taiwan society. The Tan-Hsin Archive, on the other hand, is significant in having series of documents concerning a single event happened in the old Taiwan society. As these two collections have similar spatial and geographic coverage, the Tan-Hsin Archive could serve as supplements to the rubbings for doing research on the subjects described in the relic rubbings. However, the rubbings are mostly over-sized documents and not easy to provide for use. The Tan-Hsin Archive contains a tremendous amount of documents, therefore, it is difficult to retrieve documents concerning a definite topic by using microfilms or duplicates of the originals. Now that database of these two collections is available, it is possible to easily retrieve documents concerning the same event from these two collections. As a result, it enables cross-referencing use of these first-hand historical resources. The paper hopes to initiate new research possibilities based on the rubbings and encourage in-dept research from referencing use of Tan-Hsin Archive. In addition, the authors made an approximately corresponding topical map of these two collections from subject analysis, adding it as an appendix of this article for researchers' reference.

標題是圖書館提供目錄查詢的檢索點之一，藉由標題概念及標題款目的規範，使檢索者之資訊需求與專業人員之主題分析獲得充分溝通與協調後，能夠獲得最大的檢索效益。為了解目前中文標題能夠達成之檢索效益及可能產生的缺失，本研究以國立臺灣大學線上公用目錄系統之檢索者為研究對象，共蒐集48份有效樣本，分析其中文標題檢索行為與檢索效益，冀能對於中文標題款目結構與線上公用目錄標題檢索設計提出具體改善建議。

Subject heading is one of the access points enabling subject searchers to retrieve the best available materials in the library. This paper uses 48 collected samples concerning the Tulips system (National Taiwan University Library Information Processing System) to analyze user searching behavior and the effectiveness of retrieval under the Chinese Subject Headings. On the one hand, the paper tries to provide some suggestions to improve the structure of the Chinese Subject Headings, and on the other hand to enhance system features of online catalogues.

本文探討分類不一致對自動分類成效的影響。經由近似文件的自動偵測，以及兩種分類方法針對兩個測試文件集做的比較實驗，本文發現：訓練資料的分類不一致性，即便高達34%，幾乎也不會影響分類器的成效。此項發現，其重要的意涵是，即便過去的研究使用了一致性不高的測試集做實驗，其結論仍舊是有效的。當然，分類不一致性高的資料，拿來訓練後，不管分類器好壞，其得到的分類成效都是比較低的。除了以上發現外，本文也介紹了一套中文分類測試集，免費提供各界研究使用。另外，作者也提出了一套偵測複本或相似文件的可靠方法，與過去的方法比較，此方法可以偵測過去方法所無法偵測到的相似文件。

This article discusses the effect of inconsistency in training data on the performance of text classifiers. Our experiments show that the inconsistency, even reaching a level as high as 34%,hardly affects the effectiveness of he classifiers.Better classifiers perform better independent of duplicates and label inconsistency.The implication is that past experiments(especially on the Reuters-21578 collection) remain valid. In the experiment process,the author proposes a duplicate detection technique that is far more effective than previous ones.A new Chinese test collection for text categorization is also introduced for deneral free download.