隨著網際網路(Internet)的資訊蓬勃發展，使用者可以輕易的從各個搜尋引擎與入口網站取得大量的資訊。然而，在此同時，使用者也得面對資訊過載(Information Overload)的問題，資訊過濾(Information filtering)也就應運而生。然而，使用者的興趣並非一成不變，它會隨著時空的變化產生改變。這種目標概念隨著時空改變而轉變的現象稱之為概念飄移(Concept drift)。以往的研究多關注在單標籤分類(Single label classification)所發生的概念飄移，然而現實生活上使用者對於資訊的需求是多元、多主題的，並且每個主題在時空的影響下擁有各自的喜好變化；同時文件也常屬於多個類別，若僅依照文件的主要概念，將之分類，則可能讓使用者錯過潛在感興趣的相關文件。因此本研究提出一個以字詞網路為基礎的使用者模型，透過它可以依照使用者對於多個主題的喜好對文件進行過濾，而在喜好發生變化時，也能夠適當的偵測並更新模型。 With the rapidly growing of internet, users can easily access mass information from a variety of search engines and portals. However, users also have to face the problem of “Information Overload” in the meantime. Therefore, the research of information filtering has been caused. Nevertheless, the users' interest are not static, they will change with time and space. The phenomenon that the distribution of data changes over time is called “Concept drift”. Previous researches about concept drift usually focus on the situation of single label classification. But in fact, the demand for information is diverse and user may be interested in multiple target concepts. And each concept has its own drift pattern. Furthermore, documents often belong to more than one class. People will miss potentially relevant documents if only considering the main concept in classification. Therefore, this paper proposes a keyword-network based user model, through which people can filter incoming documents according to their preference. When one of target concept has drift, the user model also has the ability to adapt this change.
