在行動社群網路中探勘使用者活動和移動行為

Abstract

隨著適地性社群網路(LBSNs)的蓬勃發展，使用者可以在這些平台上與朋友們分享打卡資訊。而從這些適地性社群網路平台所取得的打卡資料不單表示使用者所在的時間與空間，更重要的是使用者在此時空狀態下所從事的行為。藉由分析這些打卡資料，我們可以提供使用者個人化的適地性服務。此篇論文主要有兩個目標，分別為：1) 針對使用者打卡資訊中的時間與地點資訊，推測該使用者當前所從事的活動(activity)。2) 針對使用者當前所從事的活動，推測該使用者可能出現的地點(location)。我們將面臨使用者打卡筆數不足的挑戰，為解決此問題，我們利用個人化的貝氏網絡(Bayesian Network)來描述打卡資料中時間、空間，與活動這三項因素。接著，將主要問題分成兩部分，分別探討時間與活動的關係(time-activity correlation)，和地點與活動的關係(location-activity correlation)。針對時間與活動關係的部分，我們提出行為轉換模組(Activity Transition Model)來計算在特定的時間區塊中，從事每項活動的機率。針對地點與活動關係的部分，採用高斯混合模組(Gaussian Mixture Model)，為每位使用者的每項行為來建構。本篇論文中，採用一組實驗資料來驗證我們的方法，並與傳統的貝氏機率、支持向量機、非負矩陣分解等方法進行比較，針對空間的部分則是與核密度估計法(Kernel Density Estimation)比較。實驗結果證實，不論是預測使用者活動或移動行為，我們的方法都優於其他。

With the popularity of location-based social networks (LBSNs), users would like to share their check-ins to their friends for more social interactions. These check-in records reflect not only when and where they are but also what they are doing. If we can capture user activity and mobility features in LBSNs, the location social platform can provide more personalized location-based services for users. In this paper, we aim to infer individual user activity and mobility based on their check-in records in LBSNs. To infer individual activities, given a user, a location and a specific time, we propose a Bayesian-based approach to evaluate the probability of the user performing an activity at the given location and time. Moreover, to infer individual mobility, given a user, an activity and a specific time, we can also utilize the above approach to derive the location distribution of the user performing the given activity at the given time. Our proposed Bayesian-based approach contains two major components, time-activity and location-activity model. We analyze the relations between activity and time, location and activity from check-in records. Then, we utilize Gaussian mixture model and Markov based techniques to model the location-activity and activity-time relations, respectively. In the experiments, we select a real datasets, and the experimental results show that our proposed Network-based Activity Inference Model has higher accuracy than the state-of-the-art approaches for activity inference.