| 摘要: | 以學術的角度來看，資訊擷取技術(Information Retrieval, IR)主要是用於透過搜尋機制幫助使用者在圖書館找尋文件內容（Content）。在圖書館環境中，數位或實體文件是必須經過分類與整理，而每份文件內容也必須經過人工的處理，將文件的基本資訊如文件（書籍）名稱、作者、出版日期、出版商、摘要、分類類別、關鍵字詞與數位內容本身以結構化與非結構化的資料存放於資料庫中。因而資訊擷取的應用主要是依據搜尋者的關鍵字詞，在結構化與非結構化的資料庫中找到所有可能與相關的數位內容，並依其優先順序呈現。 一般搜尋數位內容的作法是利用關鍵字詞比對(Keyword Matching)或各種的相似度公式作資訊的擷取。然而採用關鍵字詞查詢或相似度公式的作法，不容易從數位圖書館中擷取到『所有相關的資料』，原因是人們使用自然語言作表逹，用不同的詞彙表達本身要傳逹的概念，致使以關鍵字詞比對的召回率(Recall)無法提升。本篇論文提出的主題概念階層模型(Topic/ Domain Concept-Hierarchy Model)是將領域知識概念作階層式的分類，形成一個領域概念階層。每個分類項為一個概念節點，其對應相關的文件集。概念節點關鍵字詞來自於文件集，在領域階層那些概念節點中，搜尋使用者所下的關鍵字詞，若二者符合，表示搜尋者想了解此概念節點的內容，此節點稱為相關概念節點（Relevant Conceptual Node），而以下的節點稱為相關概念子節點( Relevant Conceptual Subnode)，利用五個變動因素：節點之階層數,使用者的關鍵字落於節點之個數,使用者關鍵字詞與相關概念節點之cosine相似度,概念節點與子節點之距離,節點之分支度來調整計算相關概念節點與子節點之權數與相似度值。 實驗數據證明主題概念階層模型能有效地應用在資訊擷取，能帶出搜尋者想要的搜尋目標與其相關的數位內容，並依據使用者最合適與相關聯的優先順序作排列，在最短時間內擷取他最想要的數位內容。 In viewpoint of the academic, Information Retrieval method is used to facilitate content search in a library environment. In a library, librarian needs to establish description information of digital content or physical content before stored. The description information will stored into repository including title, authors, published date, publisher, abstract, category, terms and the contents. Therefore, retrieval process is implemented based on comparison between user’s query and repository. In general, keyword matching is a common approach in information retrieval research. However, this approach can not always brings a lot of all relevant information. The main reason cause this result is that people may use different words to access a specified information. Therefore, the recall performance of keyword-match is poor. In our study, we proposed Topic/ Domain Concept-Hierarchy Model to transform domain knowledge into hierarchical category in a domain hierarchy. Each category is a concept node and has corresponding content set. The represented keyword of node is extracted from content set. The matching is executed in the domain hierarchy to compute the similarity between user’s query and keywords in domain hierarchy. If matched, it means user intend to browse corresponding content set. The Node is call relevant conceptual node (RCN) and its bellow nodes are relevant conceptual sub-node (RCS). Experiment result shows the proposed Topic/ Domain Concept-Hierarchy Model can be applied to information retrieval effectively. The recall and precision has been significantly improved comparison with traditional method. The responded result is ranked followed the correlation in domain hierarchy. In this way, users can retrieval suitable material in a short time. |
