Mining Application Usage Patterns of Smartphone Users

近幾年內，行動裝置科技、尤其是智慧型手機的蓬勃發展，在許多方面都影響了人們工作和生活的方式。具備不同功能的應用程式，能夠增強智慧型手機的功能，並且更加符合使用者的個人需求，為人們帶來許多便利和娛樂。因此，了解使用者使用何種應用程式、以及使用者如何使用這些應用程式，是相當重要的議題。 研究者們已經提出了許多相關的研究，以期提供更好的使用者經驗給智慧型手機使用者。然而，這類研究大多是根據問卷調查、或者在實驗室環境下所進行的實驗，只能有限的反映真實世界中的使用者行為。在本論文中，為了克服這些限制，實作了資料探勘的技術，應用在一長時間收集而來的真實資料上，以發掘使用者行為中的顯著規律。 首先，本論文應用了傳統的關聯式規則探勘、以及序列式規則探勘的演算法，並且針對他們的不足進行討論。接著，本論文定義一段session為使用者單次的使用區段，包含使用者在這段期間內的行為記錄，並且提出一個資料前處理的流程以轉換原始資料成session。最後，本論文提出一個序列式規則探勘演算法的延伸，藉由限制session內應用程式使用的時間間隔，來產生代表更強烈關聯性的規則，並且蘊含更豐富的智慧型手機使用者行為資訊。 本論文對於智慧型手機應用程式的使用者行為分析，做出以下三個貢獻：第一，本論文提供一套資料前處理的流程，可用於此類型的分析研究；第二，本論文提出一個從序列式規則探勘演算法所延伸而出的演算法，特別適用於智慧型手機使用者行為記錄資料的規則探勘；第三，本論文提供了規則的相關分析。 本論文使用真實世界的資料集來探勘，發掘出的規則可以反映和解釋智慧型手機使用者的行為，並且，透過詳細的分析，這些結果有助於智慧型手機中，相關應用程式以及使用者介面的設計。The development of smartphone technology has changed the way people work and live in many aspects in just a few years. The advanced operating systems on smartphones allow developers to create applications that enhance the functions of the smartphones, and the applications help users personalize their smartphones to enrich their work and lives. Various types of applications designed for smartphones emerge and bring people convenience and entertainment. Therefore, it is practically interesting and important to study what applications people use on their smartphones and how people use these applications. Many studies are conducted in order to provide a better user experience to smartphone users. However, most of these studies are based on questionnaire surveys or designed experiments that are performed in a laboratory setting and would be limited to reflect the actual application usage done by users in their daily lives. To overcome the limitation, we apply data mining techniques to a data set, which contains smartphone logs collected from several users over a long period of time, and intend to discover meaningful patterns that represent user behaviors. First of all, we apply the classic association rule mining algorithm and the traditional sequential pattern mining algorithm to the data set, and we discuss their weaknesses for mining application usage patterns of smartphone users. Then, we propose a data processing process that transforms the logs into sessions rather than transactions or sequences. We define a session as a user’s one-time usage period. Each session contains the user behavior during the period, including the applications used. Next, we propose an extension of a sequential pattern mining that allows us to take time constraints into account. Stronger rules can be generated by restricting shorter intervals between applications used, and they give us much deeper knowledge of user behaviors. In summary, this thesis contributes to user behavior analysis for smartphone applications from the following three aspects: First, it provides a data processing process that is essential for the analysis. Second, it provides an algorithm extended from a widely used sequential pattern mining algorithm and designed to discover rules from logs recording activities of smartphone users. Third, it provides an analysis on the discovered rules. This thesis contributes to a better understanding of smartphone users’ behaviors and a detailed analysis to the results. It uses a real data set to discover rules that could be beneficial to the designers of smartphone applications and user interfaces