標題: | 在智慧型手機上針對時間相關的行動應用程式做使用者行為預測 Prediction Usage Intension of Time-Dependent Mobile Applications on Smart Phones |
作者: | 沈姿柔 Shen, Tsu-Jou 彭文志 Peng, Wen-Chih 資訊科學與工程研究所 |
關鍵字: | 行動應用程式;使用者意圖;時間相關;時間概述;預測;mobile Application;user intension;time-dependent;temporal-profile;prediction |
公開日期: | 2010 |
摘要: | 隨著定位、觸控、感知技術的成熟,行動應用程式 (簡稱 App) 迅速發展以滿足各類使用者的需求。App 的使用是高度依賴於人的行為。因此,分析智慧型手機上 App 使用的紀錄,提供一個獨特的機會去發現一個人 App 使用的行為。在本文中,藉由發現智慧型手機上使用的行為,我們的目標是提供 App 使用意圖一個合理的預測,以避免用戶浪費時間尋找想要的 App。本文的目的有兩個方面:(1)從應用程序使用紀錄發現使用狀況隨時間變化的 App (稱 time-dependent),並將每個 time-dependent App 的訊息表示成一個空間概述 (稱 temporal-profile) ;(2)推導預測機制,預測一個在查詢時間上高意圖使用的 time-dependent App 集合。我們提出一個 AppNow 的框架,探索使用者 App 的使用行為,並提供在未來的時間使用者 App 使用的預測。 AppNow 由兩個階段組成,分別為時間依賴性決定階段和意圖的預測階段。在時間的依賴決定階段,我們定義時間相關的特質,以決定一個 App 的時間依賴性。為了做 App 使用意圖的預測,我們為每個 time-dependent App 產生一個 temporal-profile 來形容它的時間信息。在意圖預測的階段,我們提出兩個記分函數來評估在查詢時間 App 的使用意圖。以機率為基礎的記分函數通過各個 App 的 temporal-profile 進行預測,而以 TF-IDF 為基礎的記分函數則同時考慮多個 App 的 temporal-profile。在我們的實際資料集上,我們進行了綜合實驗。結果表明,AppNow 可以對智慧型手機使用者做有效地、準確地 App 使用的意圖預測。 With the maturations of technologies about positioning, touching, and sensing, mobile applications (abbreviated as App) are developed rapidly to satisfy kinds of users' requirements. The usages of Apps are highly dependent on human behavior. As a result, analyzing App usage logs of smart phone offers a unique opportunity to discover a person's App usage behavior. In this paper, by discovering the usage behaviors on smart phones, we aim to provide a reasonable prediction on App usage intentions to avoid users from wasting time on searching wanted Apps. The goal of this paper is two-fold: (1) discovering the time-dependent Apps from App usage logs and representing the information of each time-dependent App into a temporal-profile; (2) deriving the prediction mechanisms to predict a set of time-dependent Apps that has highly intension to be used at a query time. A framework AppNow is proposed to explore a user's App usage behavior and provide a prediction on App used by the user at future time. AppNow consists of time-dependency determining phase and intension prediction phase. In time-dependency determining phase, the time-dependent features are defined to determine the time-dependency of an App. For App usage intension prediction, we generate a temporal-profile to describe its temporal information of each time-dependent App. In intension prediction phase, two score functions are proposed for evaluating the usage intension of an App at the query time. The probability-based score function derives the usage intension by the temporal-profile of an App, while the TF-IDF-based score function does by referring the temporal-profiles of multiple Apps. Comprehensive experimental results are conducted by our real dataset. The results show that AppNow can effectively and precisely predict the usage intension of Apps of a smart phone user. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT079855505 http://hdl.handle.net/11536/48239 |
Appears in Collections: | Thesis |