以知識本體為基礎之特定領域文件段落擷取方法及應用

Abstract

本研究以發展土木及營建領域之資訊檢索技術為出發點，目的在研究過程中找出本領域文件之特徵，並嘗試以領域專業知識協助提升資訊檢索之效能。為此，本研究共分三個主要階段進行：1. 建立特定領域之測試文件集，提供一個評比檢索方法效能的平台，做為發展後續檢索方法的測試基準；2. 由特定領域資訊手冊擷取該領域之知識本體(Ontology)，透過權重分析法協助修訂以減輕領域專家投入發展知識本體之工作負荷；3. 以知識本體做為領域知識之來源，研究如何以知識本體協助擷取文件段落(Passage)，以文件段落檢索的方式提升特定領域文件檢索之效能，本方法命名為OntoPassage。本論文中將詳述各階段的研究動機、問題發想以及解決方案，一步步建構完成最後文件段落檢索所需的各項資源，除了以實驗的數據佐證OntoPassage可確實提升文件檢索之效能，亦由OntoPassage檢索系統的實作展示OntoPassage具有領域知識特徵之優點。

The major purpose of this research is to discover the information retrieval related characteristics of the domain-specifics text documents during the development of the IR models for architecture, engineering and construction domain. The authors proposed a set of research progress to finding the solution to improve the retrieval performance for domain-specific documents, including: a. preparing the testing reference collection to provide a standard platform for evaluating the developed IR models; b. propose a methodology to extract the domain ontology from domain handbooks, and to reduce the work load for participating experts while editing and revising the ontology; c. propose a novel passage retrieval model, named “OntoPassage”, it takes domain ontology to producing concept-oriented passages that could improve the retrieval performance of the domain-specific IR. In the thesis, the research objectives, problems faced and proposed solutions of each progress were discussed separated in independent chapters. The experiment results showed the OntoPassage could highly improve the retrieval performance. OntoPassage retrieval system also demonstrates the potential success of learning and sharing the domain knowledge by using the concept-oriented retrieval scenario.