| Abstract: | 隨著智慧型手機日益普及，相關研究變得更加重要；智慧型手機應用程式的數量正在 快速地增加，而這使得應用程式使用的自由度增加但卻讓相關研究變得更有挑戰性。 本計畫是目前計畫『智慧型手機應用程式使用模式探勘』(102-2221-E-004-013)的延 伸，其目的是發展用於探勘應用程式使用模式的技術，而其重點將會放在如何從日誌 紀錄當中挖掘出情境資訊。在沒有任何假設或監督的情形之下，目前計畫主要考慮使 用者使用應用程式的組合與順序，而本計畫將主要考慮組合與順序是如何隨時間變 化。現有許多研究是關於預測使用者下一個會去的地點或是下一個會用的應用程式， 但是本計畫的主要工作項目並不是發展預測模型。目前計畫的主要工作是發展用於探 勘機應用程式的使用模式，而這種模式是描述哪些應用程式常被用在一段不定長的時 間(也就是一次的使用歷程)之內。本計畫的主要工作項目是發展可用於探勘這些使用 模式的時間變化。目前計畫的最終目標是要用資料探勘技術去發現使用者的習慣，而 本計畫的最終目標將是發展技術以便發現使用者的習慣的改變。 本計畫將從以下兩個方向延伸目前的計畫：(1) 資料：本計畫將參考目前計畫所使用 的真實資料來源，但會使用更大的資料集。為了交互參照，本計畫將額外參考其他網 路上可取得的資料集。依據目前計畫所得到經驗，本計畫將花更多心力進行資料前置 處理，特別是針對地理位置資料。(2) 探勘：本計畫考慮 “動態的資料” 以及 “模式 的動態”，要將現有計畫所探討的 “靜態探勘” 延伸到 “動態探勘”。更具體地說，本計 畫預計發展在日誌紀錄資料串流上的模式探勘演算法，特別是針對描述使用者在一次 的使用歷程內連續地使用哪些應用程式的序列模式。本計畫預計發展基於時間視窗的 方法。本計畫用的原始資料集的維度不高，但資料集內有超過一千個應用程式，而每 個應用程式有可以有數個行為(功能)。每個應用程式可以是資訊的消費者和生產者， 還可以是其使用者和環境之間的介面(例如，照相和錄音應用程式)；每個使用者都可 以任意安裝、移除，以及使用應用程式，還可任意在應用程式之間切換，而不用受限 於預先設定好的 “路徑” (例如，超連結引導網頁之間的切換)。因為這些特性，一般 的網頁點擊串流探勘技術無法直接使用於本計畫。當然，它們會是很好的參考對象。 資料探勘計畫常會邀請其他領域的學者去評估或解讀探勘結果。本計畫則是不同。由 本計畫所發展的技術給出的探勘結果，將可協助社會科學學者制定研究問題(例如，“為 什麼使用者要用這種方式使用這個應用程式？”或是 “使用者特質會如何影響使用者 他們使用應用程式的行為？”)，探討應用程式的使用情境。換句話說，本計畫的產出 將可推動更多相關於智慧型手機應用程式使用的研究。
With the growing popularity of smartphones, the relevant studies become more important; the number of smartphone applications (Apps) quickly increases, and this increases the degree of freedom of App usage but makes the relevant studies more challenging. This project is an extension of the current project “Pattern Mining for Smartphone Application Usage” (102-2221-E-004-013), while its purpose is to develop techniques used to mine App usage patterns and its focus will be on how to mine contextual information from log data. Without any hypothesis or supervision, the current project mainly considers the combinations and sequences of Apps that users use, while this project will mainly consider how the combinations or sequences change over time. There exist many studies regarding the prediction of the location that a user will visit next or the App that a user will use next, but the main task of this project is not to develop prediction models. The main task of the current project is the development of techniques used to mine App usage patterns, and such a pattern describes what Apps are often used in a variable period of time (that is, a usage transaction). The main task of this project is to develop techniques that can be used to mine temporal changes of these patterns. The ultimate goal of the current project is to use data mining techniques to discover users’ habits, while the ultimate goal of this project is to develop techniques to discover the changes of users’ habits. This project will extend the current project in the following two directions: (1) Data: This project will refer to the real-world data source used by the current project, but it will use a larger data set. For cross-reference, this project will additionally refer to other data sets that are available on the Internet. According to the lessons learned from the current project, this project will spend more effort performing data pre-processing, especially for the geolocation data.(2) Mining: This project considers “dynamic data” and dynamics of patterns”, and it will extend “static mining” studied in the current project to “dynamic mining”. More specifically, this project plans to develop algorithms for pattern mining over log data streams, especially for sequential patterns each of which describes what Apps used one after another by a user in a usage transaction. This project plans to develop a time window based method. The dimension of the raw data set used in this project is not high, but there are more than one thousand Apps in the data set and each App could have several activities (functions). Every App can be an information consumer and producer, and it can be in interface between its user and the environment (for example, recorder and camera Apps); every user can arbitrarily install, uninstall, and use Apps, and he or she can arbitrarily switch between Apps without being restricted by the pre-defined “paths” (for example, hyperlinks guide switching between web pages). Due to these characteristics, general web clickstream mining techniques cannot be directly applied to this project. Of course, they would be very good references. Data mining projects often invite researchers in other fields to evaluate or interpret the mining results. This project is different. The mining results given by the techniques developed in this project will be able to assist social scientists in asking research questions (for example, “Why would the users use this App in this way?” or “How do users’ characteristics affect their App usage behaviors?”) and studying the context in which Apps are used. In other words, the output of this project would be able to drive more studies relevant to the use of smartphone applications. |
