在現今社會中，學術成果越來越受重視並被需要；然而，隨著科學成果逐年地累積，並於網際網路上公開和流傳，使用者要能夠在這些大量資料中找到符合需求的文件成為一項巨大挑戰。由於科學文件是具有結構性的文本，當中必然包含能夠用以提升使用者檢索成效的因子。本研究針對科學文件的三項特徵：字型、位置和引用文獻等加以探討，這三項特徵在過去雖然有個別的文獻探討，但截至目前未有整合三者以提升檢索成效的相關研究。首先，我們將釐清字型、位置和引用文獻這三項因子彼此之間的關係，之後將依據其存在關係，透過結合三者來設計出能夠用以提升文件檢索成效的方法，最後，我們經由真實科學文件進行實驗，來實證本研究方法的有效性和成效。 As the fast dissemination of research results on the worldwide web, a user’s task of finding useful information becomes more challenging. Usage of scholarly material is growing rapidly and there is a growing demand for high-quality scholarly information. Since a scientific document is a structural text, there would have some useful features that can be used to improve retrieval performance. Here, we investigate three features, fonts, positions and cited references. Although in the past these three individual features have been used in document search, no existing research discusses how to integrate these three together to improve retrieval performance. Therefore, we will first investigate the relationships among them, and then study how to combine them to design a novel retrieval method based on their relationships. Finally, extensive experiments have been carried out through real scientific documents to show its usefulness and performance.
