Browsing by Author "Zhou, Qian"

Energy plays a central role in mobile computing, especially energy-intensive activities such as watching videos or playing games on mobile devices have increased in popularity. These activities accelerate energy usage in the device, as a result, the question of economizing the energy consumption on mobile devices becomes relevant. Some research efforts have focused on energy management applications to prolong battery life by detecting energy-hungry applications and recommending users to close those applications. However, the recommended applications could be uniquely important to users’ mobile experience and usage might continue even if it means decreased battery life. Except increase battery life by economizing mobile behavior, it is relevant for the design of energy-saving applications to know how users behave when receiving both helpful and redundant recommendations. We conduct a study on mobile application user behavior when there is a mobile energy-aware application (Carat) present on the devices. This thesis provides an approach by using application usage as implicit feedback to study if user behavior changes when recommendations on energy-hungry applications are given over the study period. Firstly, the thesis describes procedures for pre-processing and cleaning the study datasets, such as running applications in sample dataset and energy-hungry applications recommended by Carat in bug dataset and hog dataset. Secondly, this thesis provides statistical analysis methods for analyzing mobile data in different aspects. For example, applications are divided into system and installable applications. We found that users have more common system applications on their devices while less overlapped installable applications. We also separately study bugs and hogs which are the two types of energy-hungry applications. In general, there are more unique energy-hungry applications detected as hogs than bugs. For an average user, system applications are slightly more often bugs than installable applications while installable applications are more often hogs when compared with system applications. Thirdly, this thesis utilizes point biserial correlation to study application usage and Carat recommendations. We found there is no relationship between application usage and recommended energy-hungry applications. We also found that Carat users previously collected information to make recommendations. In addition, we found applications might needed by users. Based on our findings, we suggest that Carat and other energy-hungry applications recommend actions based on recent data only, and do not recommend actions against user’s needs. ACM Computing Classification System (CCS): General and reference → Cross-computing tools and techniques → Empirical studies Probability and statistics → Statistical paradigms → Exploratory data analysis Human-centered computing → Human computer interaction → Empirical studies in HCI