HiNextApp: A Context-Aware and Adaptive Framework for App
Prediction in Mobile Systems
Chaoneng Xiang, Duo Liu
∗
, Shiming Li, Xiao Zhu, Yang Li, Jinting Ren and Liang Liang
§
∗
College of Computer Science, Chongqing University, China
§
College of Communication Engineering, Chongqing University, China
Abstract—A variety of applications (App) installed on mobile
systems such as smartphones enrich our lives, but make it more
difficult to the system management. For example, finding the
specific Apps becomes more inconvenient due to more Apps
installed on smartphones, and App response time could become
longer because of the gap between more, larger Apps and
limited memory capacity. Recent work has proposed several
methods of predicting next used Apps (here in after app-
prediction) to solve the issues, but faces the problems of the
low prediction accuracy and high training costs. Especially,
applying app-prediction to memory management (such as
LMK) and App prelaunching has high requirements for the
prediction accuracy and training costs.
In this paper, we propose an app-prediction framework,
named HiNextApp, to improve the app-prediction accuracy and
reduce training costs in mobile systems. HiNextApp is based on
contextual information, and can adjust the size of prediction
periods adaptively. The framework mainly consists of two
parts: non-uniform bayes model and an elastic algorithm.
The experimental results show that HiNextApp can effectively
improve the prediction accuracy and reduce training times.
Besides, compared with traditional bayes model, the overhead
of our framework is relatively low.
Keywords-App prediction; contextual information; mobile
systems;
I. INTRODUCTION
Mobile Apps installed on mobile systems are becoming
more and larger, thus posting more pressure on energy
consumption and memory demand. Accurate app-prediction
makes it feasible for mobile systems to acquire user be-
haviors and to improve mobile user satisfaction. However,
there exist two key problems to be solved for app-prediction:
low prediction accuracy and high training cost. Especially
in some application fields of app-prediction such as memory
management and App prelaunching, high prediction accu-
racy and low training costs are required. Low prediction
accuracy is mainly due to unstable and changing smartphone
user behaviors [1]. High training cost is mainly due to
the utilizing of a large number of different sensor data in
training periods, which increases energy consumption and
memory demand [2]. In this paper, we focus on improving
the accuracy of app-prediction and reducing the training cost
at the same time.
Recent work has explored a variety of app-prediction
problems including: (1) proposing several app-prediction
methods [3], [4], [5], [6], [7], [8]. A Bayesian network
model and a linear model are proposed to predict next used
Apps [6]. The work in [7] and [8] focuses on selecting
∗
Corresponding author: Duo Liu, College of Computer Science, Chongqing
University, Chongqing, China. E-mail: liuduo@cqu.edu.cn.
appropriate contextual information to improve the app-
prediction accuracy. (2) reducing the cost of training the
app-prediction model [2], [9], [10]. A temporal-based Apps
predictor based on temporal information is proposed to save
the energy consumption [2]. Chon et al. propose an efficient
technique to maximize the accuracy with a given energy
constraint [9]. One personalized feature selection algorithm
which can successfully reduce the log size and the prediction
time is proposed [10]. (3) solving the cold start problems.
The strategy using data sets of similar users is proposed
to solve the problems [11]. (4) applying app-prediction to
different fields [12], [13], [14], [15], [16], [17], [18], [19],
[20], [21], [22], such as finding a specified App from a mass
of installed Apps more easily, reducing perceived delay of
smartphones, improving memory management and so on.
This work falls in the first two categories. Generally,
the previous work about app-prediction based on traditional
bayes model has achieved better prediction accuracy, but
ignores the two essential characteristics of app-prediction:
(1) the recent records in training data are more helpful to
app-prediction than the ancient ones [4]. In the daily use
of smartphone, users would install new Apps sometimes,
thus their App usage behavior would change. As a result,
App usage pattern may change every now and then, and the
ancient records would be too out-of-date compared to the
current App usage pattern. (2) more training data does not
mean higher prediction accuracy [4]. This is because use-
less (out-of-date) training data would reduce the prediction
accuracy of newly installed Apps. Therefore, in this paper,
we take the characteristics into account in traditional bayes
model. Besides, we also consider to reduce training costs
in our proposed app-prediction framework. Due to the gap
between high training costs of deep learning methods and
the limited computing resources on smartphones, we do not
take deep learning methods into account in this paper.
In this paper, we propose a context-aware app-prediction
framework, called HiNextApp to effectively improve the
prediction accuracy and reduce the training cost by de-
creasing the training times. The basic idea is to assign
the recent records with larger weights, to adjust the size
of prediction periods dynamically. To achieve the goal, we
design an app-prediction framework HiNextApp consisting
of non-uniform bayes model and the elastic algorithm. In
non-uniform bayes model, we adopt different weights to
achieve different treatment for records in training data. It
is helpful to improve the app-prediction accuracy. And the
elastic algorithm focuses on reducing the training times
dynamically by adjusting the size of prediction periods
nearly without reducing the prediction accuracy.
2017 IEEE Trustcom/BigDataSE/ICESS
2324-9013/17 $31.00 © 2017 IEEE
DOI 10.1109/Trustcom/BigDataSE/ICESS.2017.312
776