A Context Quantification Method Based on Mobile User Behaviors
Yancui Shi, Congcong Xiong, Jucheng Yang, Yarui Chen
Institute of Computer Science and Information Engineering
Tianjin University of Science & Technology, Tianjin China
syc@tust.edu.cn
Abstract
In order to improve the accuracy of learned mobile user
preferences, a context quantification method is proposed.
Firstly, it needs to judge whether the mobile user
preference is affected by the context through analyzing the
mobile user behaviors and calculate the impact of context
towards the mobile user preference. Secondly, the
weighted association rules is employed to mine the
relevancy between the contexts. And then, the context
similarity is calculated by analyzing the mobile user
behaviors under the given context. Finally, integrating the
context relevancy and the context similarity, a context
quantification method is proposed. The experimental
results show that the proposed method surpasses the
existing methods in the accuracy.
Keywords-context quantification; context relevancy; context
similarity; mobile user behavior
I. Introduction
The mobile user preference(behavior) will be affected by the
around context. In order to learn the mobile user preference
correctly and meet the personalized demand of the mobile user,
the researchers introduced the context into the learning model
of mobile user preference [1]. However, how to introduce the
context into the classfication reasonably is a urgent problem to
solve when the classification is employed to learn the mobile
user preference.
It was shown in [2], the context was fallen into three
representation methods: scale context, ordinal context and
categorical context. The scale context: time (hours, days,
weeks); the ordinal context: temperature (15°C), age (20); the
categorical context: location (at home), mood (sad, happy). For
the ordinal context, the closer the values of the contexts are, the
more similar the two contexts are. The ordinal context can be
directly applied into the classification and the calculation of
context similarity, so it doesn’t need to quantify. But the scale
context and the categorical context are difficult for the
contextual mobile user preference learning. There is no unified
approach about how to introduce the scale context and the
categorical context into the classfication reasonably. Some
researchers quantified the context according to their own
habits. As shown in [3], Dao et al. Gave the rules to quantify
the context, e.g., 1=workday (Monday-Friday), 2=weekend
(Saturday-Sunday). Oku et al. [4] gave the similar rules, e.g.,
0=Monday to 6=Sunday. According to the method in [3], the
obtained similarity between Saturday and Sunday is 1, while
the similarity between Saturday and Sunday is less than 1
obtained by the method in [4]. Obviously, the above methods
are unreasonable and not universal.
In addition, the mobile user preference may be affected by
several contexts, and there may be some relevancy between the
contexts. Considering the above factors, the quantification
method which integrates the context similarity and the context
relevancy is proposed by analyzing the mobile user behaviors in
this paper.
The paper is structured as follows: Section II introduces the
proposed method, including the determination of the context,
the calculation of context relevancy and context similarity, and
the quantification of the context. In Section III, we do the
experiment on the real dataset and give the corresponding
analysis. Finally, we draw the conclusion in Section IV.
II. The Proposed Method
In this section, we will elaborate the proposed method. The
method includes the determination of contexts which influence
mobile user behaviors, the calculation of the context relevancy
and the context similarity, and the quantification of context.
A. The determination of context
In the paper, the volatility of mobile user behaviors is
employed to judge whether the mobile user preference is
affected by the given context [5]. The volatility is represented
as follows:
lji
lji
llq
lqji
jil
Csu
Csu
Cc
csu
suC
Vol
VolVol
vol
,,
,,,,
,,
||
(1)
Where, u
i
∈U represents the mobile user i, U represents the set
of mobile users; s
j
∈S represents the mobile network service j
(e.g.: game, music, video), S represents the set of mobile
network services; C
l
represents some context, such as time,
location etc; c
lq
∈C
l
represents the specific context instance,
e.g.: morning, noon, afternoon, evening; n
l
represents the
number of context instances that C
l
contains. n
l
will change
when the different context partition methods are used.
Introduced by the case of the time context, if it is divided into
morning, afternoon, evening, night, n
l
=4; if it is divided into
hour, n
l
=24.
represents the usage volume that u
i
used
s
j
under c
lq
. The units of the usage volume of different mobile
network services are different, such as duration, time, flow etc.
In the paper, the formula of usage volume is as follows:
lqi
lqji
lqi
lqji
lqji
cu
csu
cu
csu
csu
N
N
L
L
Vol
,
,,
,
,,
,,
*)1(*