A Preliminary Study on Applying Text Categorization to Medical Passage Retrieval

以網際網路來尋找醫學資訊之相關資訊與知識，已漸成為大眾化之途徑，使用者常需於大量資訊中尋覓具有特定主題或概念之資訊片段，故醫學資訊片段檢索系統相當重要。本文實作並分析常見之三種資訊片段檢索方法，發現此三種以字詞頻率為主要考量的方法於實務上尚有待改良之處。所以我們揭示一個以語意為主要考量的方法，探究文件分類在醫學資訊片段檢索上之可能貢獻，並探討其與其他方法結合之方式，期能獲得兼備字詞頻率及語意考量之醫學資訊片段檢索系統，做為醫療決策支援與醫學資訊探索之用。
The Internet has been a popular platform on which users find medical information and knowledge. In the huge information space on the Internet, users usually need to seek passages that convey specific topics or concepts. Medical passage retrieval is thus essential. In this paper, we implement and analyze three popular passage retrieval methods, which rank passages on term frequency computation. We identify their weaknesses in supporting medical passage retrieval, and accordingly propose a framework that employs text categorization to rank passages. We also discuss several ways to combine the framework with other methods, so that the system may consider both term frequencies and semantics of passages. The contribution is significant to the development of medical passage retrieval systems to support healthcare decision making and medical information exploration.