線上書籤社群網站使用者之興趣變遷分析

Abstract

線上書籤社群網站是一種新興的網站類型，結合了免費線上書籤服務與Web2.0的特性，主要提供儲存書籤、分享書籤、讓使用者由社群網路發掘新資訊三種功能。 當使用者將網頁存進線上書籤並以標籤去標記，我們認為此人對於網頁隱含的主題感興趣，而且用來標記的標籤是對此主題的概念性描述。在本研究中，標籤會被轉換為概念，我們將透過概念與概念在不同時間的相關度，去分析整體使用者興趣的變遷。 本篇論文中提出四種方法估算兩個概念之間的時序性相關度，以及四種不考量時間性的靜態性相關度估算法。然後我們以線性迴歸模型去偵測時序性相關度序列中是否有趨勢存在，再以偵測到的趨勢和靜態性相關度去估算概念之間的動態性相關度。 我們將偵測出的趨勢圖形化以呈現使用者興趣的變遷情形，並以標籤預測實驗評估各種方法的效能。最後討論實驗結果，提出未來研究的方向。

Social bookmarking website is a new kind of site which combined the online bookmark service and the characteristics of Web 2.0, allows users to save bookmarks, share bookmarks and explore new information in the social network. When a user bookmarks one page and annotates it with tags, we believe that the user is interested in the implied topic of this page, and the tags which were used are conceptual descriptions of the topic. In this work, all tags be preprocessed and transformed into concepts. We try to analyze the evolution of user's interest by estimating the correlation between concepts over time. In this thesis, we introduce four methods to estimate the temporal correlation between two concepts and four methods to estimate the static correlation between two concepts. Then, we employed some linear regression models to detect whether there is a trend in the history of temporal correlations. Using the detected trends and static correlations, we can estimate the dynamic concepts correlations. We visualize some detected trends to analyze the variation of users' interest, and evaluate our methods via tag prediction. Finally, we discuss our experiment results and state some issues for future work.