=Paper=
{{Paper
|id=Vol-260/paper-7
|storemode=property
|title=A New Context-Aware Learning System for Predicting Services to Users in Ubiquitous Environment
|pdfUrl=https://ceur-ws.org/Vol-260/paper07.pdf
|volume=Vol-260
|dblpUrl=https://dblp.org/rec/conf/isuvr/LeeOJ07
}}
==A New Context-Aware Learning System for Predicting Services to Users in Ubiquitous Environment==
https://ceur-ws.org/Vol-260/paper07.pdf
International Symposium on Ubiquitous VR 2007 1
A New Context-Aware Learning System
for Predicting Services to Users
in Ubiquitous Environment
Jieun Lee, Sanghoun Oh and Moongu Jeon
Abstract — This paper represents a new context-aware learning
system to provide services in ubiquitous computing environment.
The aim is to precisely decide which services each user provides.
To achieve this goal, we design a preprocessing method (i.e.,
context modeling) to obtain good information which represents
user’s characteristics from context-aware information (i.e., user
profiles) which consists of states: who, when, where, why, what,
and how: 5W1H. The proposed system applies the state-of-the-art
naïve Bayesian Decision Theory, which is one of the statistical
analyses based on probability theorem.
Figure. 1 Overall procedures of the proposed contest-aware
Index Terms — context-aware application, naïve Bayesian learning system
classifier, and user’s preference learner
The paper is organized as follows. In section 2, we
introduce the proposed context-aware learning system. The
I. INTRODUCTION paper concludes in section 3.
R ECENTLY, many context-aware applications (CAAs) have
been developing to provide various services to each user in
ubiquitous computing environment. CAAs help each user II. PROPOSED CONTEXT-AWARE LEARNING SYSTEM
to decide his or her decision as extracting meaningful This research is the sub-system of user profile management
information from many kinds of contexts such as locations, system which plays a role in managing all data for
identifications, activities, states of users, and so on. These personalizing users [1]. A proposed context-aware learning
services apply most of the real world entertainments such as system is to provide high quality services to each user as
health, education, accident, rescue, and shopping. automatically learning based on user’s preference. Figure. 1
To provide more efficient services, most of the CAAs shows overall procedures of the proposed context-aware
have tried to employ machine learning (ML) techniques with learning system which is classified by two parts: context
learning data. From this integration of context-aware and ML modeling and preference learner.
technologies, we can configure artificial intelligent service First, the context modeling processes context-information
systems. from data collected by many kinds of sensors, time, and user
In this paper, we propose a new context-aware learning preference [2]. Table. 1 describes items and types of user
system using a naïve Bayesian classifier which is one of the profiles using in the proposed system.
most popular ML methods. The main goal of the proposed
system automatically recommends suitable services to each
user. To learn this system, we employ 6 kinds of contexts (i.e.,
user profile: who, when, where, why, what, and how (5W1H))
[1].
F. Author is with the Gwangju Institute of Science and Technology (GIST),
MS student, Korea (corresponding author to provide phone: 82-62-970-2410;
fax: 82-62-970-2384; e-mail: jieun@gist.ac.kr)
S. Author is with the Gwangju Institute of Science and Technology (GIST),
Ph. D. student, Korea (corresponding author to provide phone:
82-62-970-2410; fax: 82-62-970-2384; e-mail: oosshoun@gist.ac.kr).
T. Author is with the Gwangju Institute of Science and Technology (GIST),
professor in Mechatronics, Korea (corresponding author to provide phone: Table. 1 Data types and values
82-62-970-2406; fax: 82-62-970-2384; e-mail: mgjeon@gist.ac.kr).
International Symposium on Ubiquitous VR 2007 2
[2] Andreas Krause, Asim Smailagic, and Daniel P. Siewiorek,
"Context-Aware Mobile Computing: Learning Context-Dependent
The second part is preference learner for users. This
Personal Preferences from a Wearable Sensor Array," IEEE Transactions
process consists of a rule-based engine and a naïve Bayesian on Mobile Computing, vol. 05, no. 2, pp. 113-127, Feb., 2006.
classifier. The rule-based engine generates (Rule Constructor), [3] Nurmi, Petteri, "Bayesian classifiers for context-aware computing,"
saves, and manages rules (Experience) [4]. Especially, this Research Themes in Context-Aware Computing - seminar, Department of
engine makes new rules in order to apply some information not Computer Science, University of Helsinki, Finland, January 2004.
[4] Hani Hagras, Victor Callaghan, Martin Colley, Graham Clarke, Anthony
represented by user profile into learning scheme. The primary Pounds-Cornish, and Hakan Duman, “Creating an ambient-intelligence
role of the naïve Bayesian classifier is to provide the environment using embedded agents,” IEEE Intelligent Systems
appropriate services to users using those rules. This Magazine. v19 i6. 12-19.
classification system is useful in our learning system based on [5] J.-H. Hong and S.-B. Cho, "기계학습과 지능형 에이전트(인공비서),"
정보과학회지, vol. 25, no. 3, pp. 64-69, 2007.
user’s preference because naïve Bayesian classifier is the
simplest form of Bayesian networks and is able to easily
calculate probability as assuming whole attributes in
context-information is independent [3]. In the proposed system,
whole data are categorized by 6 items considering dependency
between each attribute in the context modeling process;
therefore, these attributes is independent. For example, when
daddy turn a television on in the morning, there is not a
correlation among attributes (i.e., who (Dad) and when
(Morning), or who and what service (TV.on). Therefore, 6
items of context-information are independent in each other.
However, there are 3 constraints in the proposed
context-aware learning system. Firstly, we should control the
number of training data due to limitation of resources. Next, we
should prevent needless update because of efficient
computation time. In the above constraint, we divide 3 learning
modes [4] [5]: Monitoring mode (Collecting user information),
Control mode (Considering new and changed information), and
Stable mode (Providing services until users satisfy). All of the
modes are classified by reliability which is settled by the
number of training data and variation of rules. Finally, we
define top priorities (emergency, disasters) [5].
III. CONCLUSIONS
In this paper, we have introduced a context-aware learning
system to recommend services into each user in ubiquitous
environment. The proposed system is divided by two parts: the
context modeling process and the naïve Bayesian classification
process to predict services for extracting high level information
from low level data. In the future, we will apply various
statistical machine learning methods such as advanced naïve
Bayesian methods (e.g., selective Bayesian classifier, a
tree-augmented naïve Bayesian classifier (TAN), etc) and
Bayesian neural networks to improve the performance for
providing highly qualitative services [4]. Also, in the procedure
of context modeling, we will study correlations among contexts
for the personalization of services.
ACKNOWLEDGMENT
This research is supported by the UCN Project, the MCI 21st
Century Frontier R&D Program in Korea
REFERENCES
[1] Youngjung Suh and W.Woo, “User Profile Management for Personalized
Service in Smart Environments,” KHCI, pp. 672-677, 2006.