Learning Mobile App Usage Routine through Learning Automata

📝 Abstract

Since its conception, smart app market has grown exponentially. Success in the app market depends on many factors among which the quality of the app is a significant contributor, such as energy use. Nevertheless, smartphones, as a subset of mobile computing devices. inherit the limited power resource constraint. Therefore, there is a challenge of maintaining the resource while increasing the target app quality. This paper introduces Learning Automata (LA) as an online learning method to learn and predict the app usage routines of the users. Such prediction can leverage the app cache functionality of the operating system and thus (i) decreases app launch time and (ii) preserve battery. Our algorithm, which is an online learning approach, temporally updates and improves the internal states of itself. In particular, it learns the transition probabilities between app launching. Each App launching instance updates the transition probabilities related to that App, and this will result in improving the prediction. We benefit from a real-world lifelogging dataset and our experimental results show considerable success with respect to the two baseline methods that are used currently for smartphone app prediction approaches.

💡 Analysis

Since its conception, smart app market has grown exponentially. Success in the app market depends on many factors among which the quality of the app is a significant contributor, such as energy use. Nevertheless, smartphones, as a subset of mobile computing devices. inherit the limited power resource constraint. Therefore, there is a challenge of maintaining the resource while increasing the target app quality. This paper introduces Learning Automata (LA) as an online learning method to learn and predict the app usage routines of the users. Such prediction can leverage the app cache functionality of the operating system and thus (i) decreases app launch time and (ii) preserve battery. Our algorithm, which is an online learning approach, temporally updates and improves the internal states of itself. In particular, it learns the transition probabilities between app launching. Each App launching instance updates the transition probabilities related to that App, and this will result in improving the prediction. We benefit from a real-world lifelogging dataset and our experimental results show considerable success with respect to the two baseline methods that are used currently for smartphone app prediction approaches.

📄 Content

Learning Mobile App Usage Routine through Learning Automata Ramin Rahnamoun1, Reza Rawassizadeh2, Arash Maskooki3 1Computer Engineering Department, Islamic Azad University, Central Tehran Branch, Tehran, Iran 2Computer Science Department, Dartmouth College, NH, US 3Department of Electrical and Computer Engineering, UC Riverside, CA, US r.rahnamoun@iauctb.ac.ir, rrawassizadeh@acm.org, arash.maskooki@ucr.edu

ABSTRACT Since its conception, smart app market has grown exponentially. Success in the app market depends on many factors among which the quality of the app is a significant contributor, such as energy use. Nevertheless, smartphones, as a subset of mobile computing devices. inherit the limited power resource constraint. Therefore, there is a challenge of maintaining the resource while increasing the target app quality. This paper introduces Learning Automata (LA) as an online learning method to learn and predict the app usage routines of the users. Such prediction can leverage the app cache functionality of the operating system and thus (i) decreases app launch time and (ii) preserve battery. Our algorithm, which is an online learning approach, temporally updates and improves the internal states of itself. In particular, it learns the transition probabilities between app launching. Each App launching instance updates the transition probabilities related to that App, and this will result in improving the prediction. We benefit from a real- world lifelogging dataset and our experimental results show considerable success with respect to the two baseline methods that are used currently for smartphone app prediction approaches.
Keywords Smartphone, App usage, Finite Action-set Learning Automata, Application Transition Probability Matrix

1. INTRODUCTION & BACKGROUND Since its conception, smart app market has grown exponentially. According to a report by statista website from 2014 [7], the number of Applications that are available are 1.3 million on Google’s Market and 1.2 million for Apple’s iTunes Store. There are several research works [5, 6] that show the average number of installed application (App) are more than 40 per smartphone. Similar to other mobile computing devices, smartphones suffer from limited battery power. This necessitates optimization mechanisms significant consumption contributors. One major contributor is the application load time. The launch of an App may take a few seconds [13], while the screen is on. For some resource intensive apps such as Games the application load time is even longer than average.
   One solution to reduce the app search and load time is caching apps into memory. The challenge is the large memory cost of this practice. Therefore, a learning mechanism can create a favorite app list for pre-loading into cache [12].
   App usage prediction is not a new problem and different algorithms [12] or frameworks [2, 11] are proposed to predict app usage. Some of these research works focused on the problem of dependencies between Apps launching sequence [5,13] or more specifically on transition probabilities between App usages [3]. Moreover, personalized App recommendation is another topic of research [1]. There are works that focused on the use of clustering algorithms [4, 5] to classify App usage based on contextual information. Context data can be included in sensors based activities such as running, walking or other human centric activities such as SMS, call, etc. [13, 6]. Some researches use time series model to tackle this problem [8]. Learning Automata (LA) is an old reinforcement learning method that is being used in a wide range of applications. Here we use Finite Action-set Learning Automata (FALA) [10], but there are several different models of LA are proposed in literature [9, 10].
   All these works operate based on leveraging the temporal history of app usage as input [2, 4, 8, 11]. Some approaches [4] concurrently gather other sensor information from smartphone and correlate with app launch. By analyzing the temporal history of app usage log, these algorithms create a model from user behavior over the time and use it to predict future app launching mechanism. To the best of our knowledge, most of the existing methods [1, 4, 15, 18] have the offline approach and do not perform the learning process on the device. In other words, these methods solve an offline classification problem and they classify this data in a cloud, and not on the smartphone. This characteristic imposes the burden of network reliance, privacy and response time associated with transferring the data and receiving the result. As it has been described due to size limitations of smartphones, they cannot execute computationally complex algorithms.
   Some other algorithms [15,16,17], use different sensors as data sources. Using more than one sensor for prediction provides a better accuracy in comparison to

This content is AI-processed based on ArXiv data.