本研究以社群影音網站Youtube為例,將ComScore資料庫中的家戶使用者瀏覽紀錄做為分析對象,利用使用者的期間內上站次數、期間內單次瀏覽頁數兩個變數定義個別使用者於不同期間內的黏著度狀態。本研究應用馬可夫鏈理論與層級貝式模型建立個別使用者的移轉機率矩陣,並透過建立出來的移轉機率矩陣預測使用者下三期的黏著度狀態。此外,本研究定義出六條使用者黏著度的可能遷徙路徑,以逐步迴歸法探討人口統計變數與遷徙路徑可能性之間的關係,此做法同時考慮了使用者的異質性及動態性。最後,本研究利用建立的移轉機率矩陣,計算個別使用者的移轉機率矩陣收斂狀態,因此可得個別使用者的黏著度狀態最終落點,並發現不同最終落點區隔內的使用者人口統計變數的顯著特質。透過本研究的分析,社群網站的經營者得以掌握未來使用者的黏著度狀態,進而提早擬定適當的網站經營策略。
This thesis takes Youtube, a social networking and video sharing website, as an example, conducting research by using online browser history of household users in Comscore database. We define states of user stickiness in each period by two variables, which are frequency of entering the website and pages viewed each time. We apply Markov chain theory and hierarchical Bayesian model to build up transition probability matrix of each individual user. Therefore, we can utilize the matrix to predict the states of each user in the following periods. Furthermore, this thesis sets six possible migration paths of user stickiness, and conduct stepwise regression analysis to identify the relationship between demographic variables and possibility of the migration paths, which considers both heterogeneity and dynamic of users. In the end, we use transition probability matrixes to calculate the convergence states of each individual user, and thus we can know what would be the ultimate static states of users. Also, we find that there are significant attributes of users within segments between different ultimate states. Through the research conducted, managers of social networking website can recognize future stickiness of their users, and therefore could plan appropriate management strategies in advance.