隨著使用者的閱讀習慣從紙本轉成數位、電腦轉成手機平板，使得使用者能夠隨時隨地的閱讀，不僅也增加了平均閱讀量也造成了更容易分散注意力的環境，面對這些新的挑戰，系統除了需要解決使用者興趣的概念飄移的問題以外還需要解決因網路資料規模呈指數成長而所造成系統處理即時性的問題。 而為了解決這些問題，本研究提出了使用不同居中度演算法來建立使用者模型中主題字詞圖形的核心字詞，藉由使用這些較具代表性的核心字詞使用在系統流程中能夠達到改善建立使用者模型的時間並且甚至改進了使用者模型判斷文件的效能。而在概念偏移的問題上，本研究透過多重記憶系統模型的架構於使用者模型的興趣分類上，將使用者興趣主題區分成長期與短期興趣。最後實驗證明短期興趣的動態遺忘因子能夠較快地適應興趣，而靜態的長期興趣遺忘因子能保留較多資訊。而在模擬網路串流的情況下系統的F-measure效能較以往研究高了並且提高了系統速度。 ;While user’s channels of reading is changing from physical to digital, desktop computer to mobile device. It becomes easier for user to read at anywhere, anytime. It have not only increasing the amount of average reading but also causing the user interest drift more often. To solve these problems, information filter system have to adapt the concept drift of user interests and trains fast enough to deal with the explosion of documents streaming. The research try to use different centrality algorithm to find the core set of keywords in user profile′s graph. Using the strong keywords instead of all of the keywords in the graph, system improves the speed of building user profile and even the performance of the system. In addition, the research design the user profile′s interest base on the Atkinson-Shiffrin′s multi-store model, the framework divided user interests into long-term interest and short-term interest. The short-term interest use the dynamic forgetting factor to adapt the concept drift occurred in user profile. In contrast, the long-term interest using the static forgetting factor to store information for the system to use. the experiments proved short term forgetting factor can adapt the concept drift quicker, and the long term forgetting factor can save more information in the interest. In the end, research’s system shows better F-measure performance and more efficient than the other research.
