以乏晰概念網路為基礎的中文文件擷取研究

Abstract

傳統的查詢系統,其檢索方式多以詞彙危機礎,然而 詞彙中一詞多義或一 義多詞的特性,往往造成檢索出的文件非為使用者所需.因此,本篇論文主 要是探討在乏晰概念網路中概念式檢索.在本篇論中,我們提出以乏晰概念 網路為主的擷取系統,以應用於國科會科資中心計畫報告的中文文件查詢. 在此系統中,節點間的關係值可藉由所設計的關鍵詞-類別權重函數計算得 知,同時利用乏晰集合理論中的相似函數和分群法產生新的概念.最後利用 概念矩陣模擬網路間的關係值,使其應用於文件檢索時更有效率.另外,在 本論文中,我們也提出一個以概念為基礎的文件相關查詢設計.實驗的結果 證實,所提的概念式文件相關查詢較關鍵詞式的文件相關查詢有較高的精 確度. Most of traditional retrieval systems are based on keyword search. However thefacts show that a word may contain multiple meanings and different words may contain the same meaning, then those documents retrieved by the keyword-based system may not satisfy the user's requests. In this thesis, we propose Chinese document retrieval system based on the fuzzy concept network and it is appliedto the Chinese documents of NSC. In this system, the relation values among the concept in the networks are computed by the keyword-class weighting function and new concept node will be generated by using similarity function and clustering in fuzzy set theory. Finally we use concept-matrix to implement network in order to improve retrieval efficiency. In addition, we propose a concept-based design for relevant document retrieval. The experiments show that the proposed retrieval model will produce higher retrieval precision rate than keyword-based relevant document retrieval.