摘要 現在使用者面對的資訊環境,其資訊量與資訊的增加速度都遠遠大於過去,因此考量到資訊不斷增加的資料流研究成為資訊檢索領域努力的課題,概念飄移是資料流中的資料類別隨時間改變或是使用者閱讀興趣改變造成資料篩選上的錯誤,本研究考量了系統使用者較為細膩的興趣感受,在文件相關判定階段,利用使用者看過的文件內的主題間的共現關係做為判定相關的依據,在面對資料流檢索系統的計算速度需求方面,提出了基於NGD的相似度容差方法,讓帶有相似資訊的字詞互相取來,以減少字詞數量達到降低系統執行時間的目的,而本研究將使用者的興趣分成四類,並針對四類興趣對於資料的保存與去除需求設計動態的遺忘因子,對於概念飄移發生後,系統偵測概念飄移前的時間區段造成的效能下降問題,本研究透過使用者觀看文件的行為對可能發生的概念飄移進行預測,以降低概念飄移發生時對於系統效能的影響。 ;Abstract With the amount of data and the speed of data increasing are more quickly than past time for a user nowadays. Therefore considering data stream study becomes a trend of information retrieval. The concept drift means the data categories can change by time or the data filtering mistake when user′s interests changed causing. This study considers users′ exquisite feelings, using the documents users have read belongs to which topics and judge the relevance based on the co-occurrence between two topics. The demand of the system calculating speed we propose NGD similarity tolerance method to decrease the amount of terms to reach the goal of decreasing system executing time. And our study divide users′ interests into four categories and then aim to those categories designing the forgetting factor to keep and filter the data improving the effectiveness decreasing of concept drift. This study predicts the concept drift through the users′ reading behavior to decrease the effect to the system when concept drift happened.