| Summary: | 碩士 === 國立臺灣大學 === 電機工程學研究所 === 102 === With the growing prevalence of smartphones, there is an increasing number
of mobile applications which play important roles in daily life. In this
thesis, we propose a framework of APEIC (standing for App Prediction by Environmental
and Interactional Contexts) to predict apps that are most likely
to be used according to the current context. The context consists of environmental
context (EC), which is characterized by features extracted from
built-in sensors, and interactional context (IC), which is defined as the app
launch sequence. The benefits of such prediction include fast app launching
by pre-loading the right apps into memory, and also efficient power management
by terminating apps which are not to be used in the near future. We
collected real app usage traces and made some observations that provide insights
into the design of our prediction model. First, from past traces, we
adopt features representing EC to build a naive Bayes classifier and evaluate
the launch contributions between apps from IC respectively. Second, from
the current condition, Poisson distribution is used to model the re-access pattern
of certain apps. Finally, we can rank the apps by the sum of their launch
intensities. We conduct experiments on both real data and synthetic data. The
results demonstrate the capability and the robustness of our prediction framework.
Furthermore, in combination with our companions’ work, we design a
smart launcher which helps users have rapid access to the apps they need and
recommends other useful apps according to the context at that time.
|
|---|
