CONTEXT AWARENESS OF MOBILE CONTENT DELIVERY

BASED ON FINE LOCATION ESTIMATE

Tomohisa Yamashita, Daisuke Takaoka, Noriaki Izumi

Akio Sashima, Koichi Kurumatani and Koiti Hasida

National Institute of Advanced Industrial Science and Technology (AIST)

CREST, Japan Science and Technology Agency (JST)

Sotokanda 1-18-13, Chiyoda-ku, Tokyo, 101-0021 Japan

Keywords:

Active RFID system, Context-awareness, Mobile service, Content delivery, Location estimate.

Abstract:

In this paper, to tackle with uncertainty in the real world, the light-weight ontology drive approach is proposed

for the realization of context dependent services. We concentrate on position information and an operation his-

tory, as a user’s context, and develop our location-aware content delivery system. The evaluation experiment

of our location estimate engine is performed in Akihabara Software Showcase at Information Technology Re-

search Institute. Furthermore, through the prooﬁng experiment in Expo 2005 Aichi, our proposed architecture

is conﬁrmed to enables us to realize the real world application of context dependency. Finally, we compare

our location-aware content delivery system and related researches, and discus the advantage of our system.

1 INTRODUCTION

Because of the rapid development of ubiquitous com-

puting technology, various mobile services have been

realized using personal devices, cellular phones and

IC cards, including i-mode (NTT-DoCoMo, 2007),

EZweb (KDDI, 2007), Suica (JR-EAST, 2007), and

so on. These services emphasize information access

during movement, but it is necessary to deal with rich

context of users, such as operation histories as posi-

tion information, to also provide context dependency.

In fact, development of weak radio and positioning

technologies make it difﬁcult to realize context sen-

sitive services in the real world. Therefore, research

issues of uncertainty remain among various theories

and applications.

To tackle these issues, two main approaches have

been proposed. The ﬁrst approach is realization of

pull-type services based on passive IC-tag technolo-

gies. Although this approach can achieve extremely

robust service to the uncertainty of device operation,

the data utility in context dependency has been left in

the operation of a user. Another approach is based on

the logic of probability, including Bayesian networks.

Although the introduction of various parameters en-

ables detailed tuning, time and cost issues pertain in

adjustment because of embedded logic.

From the standpoint of tackling uncertainty in the

real world, we propose a lightweight-ontology drive

approach to achieve context-dependent services. To

distinguish the observed objects and the observing

subject in a real-world application, ontologies of ar-

eas and services were developed. A mechanism of ad-

justment for the real-world application is developed in

mapping between the proposed ontologies. Through

the prooﬁng experiment in Expo 2005 Aichi (EXPO,

2005), we have conﬁrmed that our proposed architec-

ture enables realization of the real-world application

of context dependency.

2 DESIGNING CONTEXT

AWARENESS

In this study, we speciﬁcally examine position infor-

mation as a user’s context because position informa-

tion of a user shows the existence of relations with

surrounding objects tells what kind of relationship the

object has with the user. In this paper, the concept

of ”Context Awareness” is realized through develop-

ment of location-dependent content delivery. We ap-

ply an active RFID system as human location sensor

because it is only location estimate system of a large

185

Yamashita T., Takaoka D., Izumi N., Sashima A., Kurumatani K. and Hasida K. (2007).

CONTEXT AWARENESS OF MOBILE CONTENT DELIVERY BASED ON FINE LOCATION ESTIMATE.

In Proceedings of the Ninth International Conference on Enterprise Information Systems - SAIC, pages 185-190

DOI: 10.5220/0002401201850190

 SciTePress

number of visitors. However, there is no robust loca-

tion estimate system in uncertain environment.

2.1 Bridging by Linearizing

Simpliﬁcation

By supposing that we can obtain an ideal environment

in which radio wave conditions are extremely stable

in the real world, the distance from an RFID tag to

an RFID receiver is calculable based on the Received

Signal Strength (RSS) of an RFID receiver. However,

it is difﬁcult in the real world to observe the RSS pre-

cisely for the reason that the received RSS by RFID

receivers is extremely unstable even though an RFID

tag remains in a single position. Moreover, the insta-

bility of radio wave conditions is also reported as a

result of reﬂection and phasing phenomena.

To tackle real-world instability, we employ the

number of detections of a tag ID as the key parameter

for estimating its location, instead of the RSS. Re-

garding the robustness of computation, we propose a

linearizing simpliﬁcation to the relationship between

the number of detections of a tag ID by an RFID re-

ceiver (antenna) and the distance from an RFID tag

to an RFID receiver, we then regard this relationship

as ”the closer an RFID tag is located to an RFID re-

ceiver, the higher the number of detections of a tag ID

by an RFID receiver.” This linearizing simpliﬁcation

decreases the complexity of the location estimation

algorithm and enables adjustment of the parameters

in a practical period.

2.2 Lightweight Ontology

To realize a method of location estimation with the

above approximation as a computational algorithm,

we must clarify a distinction between a representa-

tion of how we recognize location and a parameter

that indicates how a computation is adjusted to the

real world. Furthermore, from the standpoint of prac-

tical use, it is necessary for the real-world application

to complete both estimating location and adjusting pa-

rameters in a very short period.

From the viewpoint of ontologies, to maintain an

adjustment of a practical application, we must distin-

guish an object that is observed and a subject that is

observing: we must also devise a means to recognize

the real world and a method to infer a location. Fur-

thermore, both of the above ontological aspects of the

location estimation must be compatible to realize a

service that is effective in the real world. The above

discussion underscores the necessity of simplifying a

model of a location estimation and a strategy of a ser-

vice that is provided. Therefore, to retain robustness

Figure 1: Aimulet GH+.

for instability of the real-world environment, we em-

ploy a hierarchical representation of areas that are rec-

ognized and contents that are serviced. At the same

time, strong restrictions and rigorous constraints are

unnecessary from the computational aspect. More-

over, correspondence between areas and contents is

considered as a parameter for adjustment. In this pa-

per, the above architecture of a hierarchical structure

including estimated locations and provided services

with fewer constraints and parameters is designated

as a lightweight ontology.

3 IMPLEMENTATION OF A

CONTENT DELIVERY

3.1 System Architecture

In this section, we explain our location-aware con-

tent delivery system (Sashima, 2004) using Aimulet

GH+, which is composed of Personal Digital Assis-

tants (PDAs) and an active RFID, as shown in Fig.

Aimulet GH+ was developed as a users’ mobile

device for our location-aware content delivery service

in Global House, Expo 2005 Aichi. The system de-

tects the location of a user with Aimulet GH+ every

second. Based on the user’s location, the system up-

dates the content list containing some items of expla-

nations about exhibits that are near the user’s location.

The user chooses one item from the content list and

touches it on the display to play an explanation with

sound, text, and graphics.

Our location-aware content delivery system com-

prises RFID antennas, RFID receivers, an RFID re-

ceiver server, a location estimate engine, and a con-

ICEIS 2007 - International Conference on Enterprise Information Systems

186

Figure 2: Data ﬂow in the location-aware content delivery

system.

tent server. The data ﬂow of the location-aware con-

tent delivery service with Aimulet GH+ in Fig. 2 is as

follows. An active RFID tag on Aimulet GH+ trans-

mits its tag ID; then RFID receivers detect the tag ID

through RFID antennas and send it to the RFID server.

The RFID server stores RFID data (RFID receiver IDs

detecting a tag ID and their respective time stamps).

The content server sends the user’s subarea to the con-

tent database, and requests a reply including the most

optimal content list.

3.2 Location Estimate

The whole area is divided conceptually into subareas.

The division method is conceptualization as layers;

some kinds of layers are prepared. Subareas in the

same layer have similar size. The sizes of subareas

in the lowest layer are smallest of all layers. Higher-

layer subareas are larger.

The location estimate engine selects one subarea

in a layer where the user with Aimulet GH+ is consid-

ered, with the highest probability, to be located. First,

the location estimate engine selects a subarea in the

lowest layer containing subareas of the smallest size.

Because the subareas in lower layers are small, se-

lection of one subarea there is more difﬁcult. Conse-

quently, the location estimate engine chooses a higher

layer containing larger subareas; it then tries to select

one subarea within that layer. Based on the number of

detections of tag IDs by RFID receivers, the selection

is processed as follows.

First, a set of layers L is deﬁned as

L = {l

,...,l

(1)

Second, set of subareas S

in layer l

is deﬁned as

= {s

1,i

2,i

,...,s

j,i

,...,s

(2)

A set of RFID receivers R is deﬁned as

R = {r

,...,r

(3)

Each RFID receiver has a receiver point. Here,

the receiver point rp

(id, T) for a tag ID id at time

T is the number of detections of tag ID id by RFID

receiver r

for one second at time T.

To consider past RFID data of t seconds ago, a

time weight of w

time

(t) is applied. With time weight

time

(t), we deﬁne the total receiver point trp

(id, T)

for a tag ID id at time T as

trp

(id, T) =

∑

t=0

time

(t)rp

(id, T −t).

(4)

Here, the time weight w

time

(t) is a monotonically de-

creasing function.

Each subarea has a subarea point. Here, subarea

point sp

(id, T) of subarea s

j,i

in layer l

for tag ID

id at time T indicates how a user with Aimulet GH+

transmitting tag ID id is considered to exist in subarea

j,i

based on receiver points around subarea s

j,i

To calculate subarea point sp

j,i

(id, T), a contri-

bution ratio is deﬁned. The contribution ratio c

j,i

)

of receiver r

to subarea s

j,i

in layer l

indicates how

the detection of RFID receiver r

contributes to infer-

ring that an RFID tag transmitting tag ID id exists in

subarea s

j,i

when RFID receiver r

detects the tag ID

id. The value of the contribution ratio is determined

based on this supposition: ”The closer an RFID tag

is located to an RFID receiver, the higher the number

of detections of an RFID tag by an RFID receiver.”

Therefore, the closer an RFID receiver is to a subarea

(or included into a subarea), the greater the contri-

bution ratio of an RFID receiver to a subarea. Each

contribution ratio is set as real number in the range

of [0,1.0]. With contribution ratio c

j,i

), subarea

point sp

j,i

(id, T) is deﬁned as

j,i

(id, T) =

∑

p=0

j,i

)trp

(id, T)

(5)

After calculation of all subarea points in a layer,

the location estimate engine selects one subarea with

the highest subarea point in all subareas in a layer.

However, if little difference exists between the high-

est subarea point sp

and the second highest subarea

point sp

, the location estimate engine select no sub-

area in layer i; instead, it selects one subarea in the

next-highest layer i+ 1. The condition by which the

location estimate engine rises from layer i to layer

i+ 1 is deﬁned as

min

ratio

i,i+1

≤ sp

/sp

(6)

Here, min

ratio

i,i+1

indicates the minimum ratio by

which the location estimate engine rises from layer i

to layer i+ 1. Otherwise, the location estimate engine

considers that a user with Aimulet GH+ transmitting

tag ID id exists in the subarea with the highest subarea

point.

CONTEXT AWARENESS OF MOBILE CONTENT DELIVERY BASED ON FINE LOCATION ESTIMATE

187

3.3 Content Delivery

The content database has a hierarchical structure of

content lists and replies with a prearranged content

list. For example, in a museum containing some exhi-

bition rooms and passages with many exhibits, in the

case that the content database receives a larger sub-

area in a higher layer, e.g. an exhibition room, the

content database responds with a prearranged content

list that includes (an explanation of) the concept of

the room above and (explanations of) the main ex-

hibits in the room below. In contrast, when the con-

tent server receives a smaller subarea in a lower layer,

e.g., the place in front of a speciﬁc exhibit, the con-

tents database replies with a prearranged contents list

that has the exhibit above and other exhibits around it,

along with the concept of the room below. However,

not many such characteristic subareas exist in all sub-

areas. Furthermore, it is difﬁcult to assign a contents

list a priori to each subarea because of the large num-

ber of subareas of all layers.

In light of the problems posed by these issues,

lightweight ontology is introduced to connect the hi-

erarchical structures of the subarea and the content

list. The hierarchical structures of the provided con-

tent list and the estimated subarea are not equivalent.

If the content lists are assigned to characteristic sub-

areas and the hierarchical structure of subareas is de-

ﬁned, then, based on the subarea’s hierarchical struc-

ture, the content database assigns the content list of

the parent subarea to the subarea to which a content

list was not assigned previously. For example, in the

case of a subarea in a middle layer, e.g., the half of

a room to which a content list was not assigned pre-

viously, the content database responds with the same

content list as its parent subarea.

3.4 Evaluation Experiment

The evaluation experiment of our location estimate

engine was performed in the Akihabara Software

Showcase (SSC) at the Information Technology Re-

search Institute (ITRI). In the Akihabara SSC, we

constructed the same system as that in the Global

House.

The Akihabara SSC area is 220 square meters

(about 15 m × 15 m). In the Akihabara SSC, 30

RFID antennas and receivers are set on the ceiling.

The whole area of the Akihabara SSC contains six

parts shown in Fig. 3, which are named based on their

facilities: lounge space, seminar room, closed meet-

ing room, open meeting room, reception, and living

room. Because the living room was under construc-

tion, it was impossible to enter it for our evaluation

Figure 3: Ground plan of Akihabara SSC.

Figure 4: Division of subareas in each layer.

experiment. Five layers were prepared for the evalua-

tion experiment, as shown in Fig. 4. Layer 4 had six

subareas. Each subarea corresponded to a room of the

Akihabara SSC in Fig. 3.

The tree structure of the subarea multi-layer sys-

tem is portrayed in Fig. 5. In this structure, the par-

ent of subarea SEACA is subarea SEAC, whose par-

ent is subarea SEA. The tree structure of the con-

tent list is also shown in Fig. 5. Here, we show the

case where the location estimate engine output sub-

area SEAC and SEACA, content database assigns a

content list R based on the content list tree structure.

The minimum ratios used in the evaluation experi-

ment are shown in Table 1.

In our evaluation experiment in the SSC, a sub-

ject with Aimulet GH+ moved around the subareas in

layer 5, and produced an actual subarea record (the

subarea in layer 5 and time in which the subject ex-

isted actually). This one comparison is deﬁned as a

single trial. The correct answers were counted, and

ICEIS 2007 - International Conference on Enterprise Information Systems

188

Figure 5: Tree structures of subareas and content lists in

multi-layer system.

Table 1: Minimum ratio in each layer.

min ratio

5,4

0.001 min ratio

4,3

0.05

min

atio

3,2

0.05 min ratio

2,1

0.1

the correct rate of each subarea was calculated.

As results of the evaluation experiment, the num-

ber of trials, the number of correct answers, and the

correct rate are listed in Table 2. From Table 2, the

correct rate of the location estimate engine in layer

4 was greater than 80% overall. In the Akihabara

SSC, trial subjects were satisﬁed with the location-

aware content delivery service using Aimulet GH+

because the provided content list was always suitable

for the subject’s location. Therefore, we conﬁrmed

that our location estimate engine had sufﬁcient accu-

racy to provide appropriate content delivery services

for users.

4 PROOFING EXPERIMENTS IN

EXPO 2005 AICHI

Expo 2005 Aichi was held in Nagoya, Japan (EXPO,

2005) from March 25 through September 25, 2005

(Duration of 185 days). The pavilions were about 90;

22 million visitors attended. Global House had three

parts: The Mammoth Laboratory , the Orange Hall,

and the Blue Hall. The Orange Hall housed a model

of the Yukagir Mammoth, Statue of Dionysus, Moon

Rock, and a model of a Voyager spacecraft.

The Japan Association for the 2005 World Expo-

sition required measures for a large number of vis-

itors, i.e., information service of exhibits to visitors

and for a reduced number of attendees. In the Or-

ange Hall, a maximum of 350 visitors entered every

20 min from 9:30 AM to 8:00 PM. Based on such a re-

quirement, we developed CONSORTS as both a con-

Table 2: Accuracy of our location estimate engine in layer

correct correct

trial answers rate

lounge space 449 406 0.904

seminar room 459 346 0.754

closed meeting room 270 239 0.885

open meeting room 280 227 0.811

reception 80 55 0.688

total 1538 1273 0.828

tent delivery system for visitors with users’ mobile

devices Aimulet GH+ and an exhibition management

support service for managers.

In Orange Hall, we provided the following exhibi-

tion management support for managers (Kurumatani,

2004), i) Detection of locations of individual visitors,

ii) Research of listening rates of contents , and iii)

Monitoring of hall congestion. As a result, we real-

ized information services for visitors and obviated the

need for about 50 attendants compared with a prelim-

inary estimate.

5 DISCUSSION

In this section, we compare our context-aware content

delivery and related studies, and discuss the advan-

tages of our system.

Recently, elemental technology for location esti-

mate has been developed and theoretical research for

spatial information has been advanced. Regarding el-

emental technology, especially that which is related

to location systems with Wi-Fi, the development of

location systems with Wi-Fi has been astonishing in

recent years; it has reached the practical use stage

(EKAHAU, 2005; AeroScout, 2005). Furthermore,

location estimate with Bluetooth, which uses Blue-

tooth RSSI and access points with variable attenua-

tors, has been developed (Bandara, 2004).

From the standpoint of comparison of ontologies,

several spatial ontologies that describe physical space

and spatial relations have been proposed, including

both geometric (e.g., GPS, GIS) and symbolic rep-

resentation (e.g., places that are identiﬁed by their

names) of space: DAML-Space, OpenCyc, SUMO,

Region Connection Calculus (RCC). A much-updated

ontology was proposed recently by the Digital Enter-

prise Research Institute (DERI) on WSMO (WSMO,

2005). However, at present, each remains as a pro-

posed description.

Moreover, services that use the ontology of loca-

CONTEXT AWARENESS OF MOBILE CONTENT DELIVERY BASED ON FINE LOCATION ESTIMATE

189

tion have been the subject of studies with similar aims

to ours (Flury, 2004; Lemmens, 2004; Chen, 2004).

However, the descriptions used in those studies rep-

resent simple recognition, e.g., ”the ﬂoor consists of

a room and a passage.” These studies did not address

performance in the real world. In another study, a

trail provides the contents of context dependence with

PDA in a real-time application in the real world (Son-

ntag, 2005). However, the clear separation of parame-

ters and ontology is not established in that study. Our

approach not only describes ”how we consider loca-

tion information.” The characteristic of our research

is mapping the structure of service: the structure of a

contents list and the structure of physical area based

on the properties of electromagnetic waves including

a parameter are shown on a map as a representation

of how we recognize location.

6 CONCLUSION

This paper presented a description of applications

of the CONSORTS architecture to an integrated ex-

hibition support system at the Orange Hall of the

Global House at Expo 2005 Aichi. Furthermore, our

location-aware content delivery system with Aimulet

GH+ on CONSORTS was taken up as a characteristic

implemented service in Orange Hall. Results of eval-

uation experiments of our location estimate engine in

Akihabara Software Showcase at Information Tech-

nology Research Institute were obtained. They will

be shown. Finally, outline of implemented services at

the Global House, Expo 2005 Aichi and further eval-

uation and analysis of our system were provided.

REFERENCES

AeroScout (2005). AeroScout, http://www.

aeroscout.co.jp/.

Bandara, U., H. M. I. M. M. H. A. T. (2004). Design and

implementation of a bluetooth signal strength based

location sensing system. In IEEE Radio and Wireless

Conference (RAWCON 2004), pages 319–322.

Chen, H., F. T. J. A. (2004). An ontology for context-

aware pervasive computing environments. In Journal

of Knowledge Engineering Review, volume 18, pages

197–207.

EKAHAU (2005). EKAHAU - Positioning Engine 4.0,

http://www.ekahau.com/.

EXPO (2005). EXPO 2005 AICHI JAPAN,

http://www.expo2005.or.jp.

Flury, T., P. G. R. F. (2004). Owl-based location ontology

for context-aware services. In Artiﬁcial Intelligence in

Mobile Systems.

JR-EAST (2007). Super Urban Intelligent Card

(Suica) of East Japan Railway Company (JR-EAST),

http://www.jreast.co.jp/e/index.html/.

KDDI (2007). EZweb, http://www.au.kddi.com/ en-

glish/ezweb/ index.html.

Kurumatani, K. (2004). Mass user support by social co-

ordination among citizens in a real environment. In

Multi-Agent for Mass User Support, Post-Proceedings

of MAMUS-03, Lecture Notes in Artiﬁcial Intelligence

3012, volume 3012, pages 1–17. Springer Verlag.

Lemmens, R., d. V. M. (2004). Semantic description of lo-

cation based web services using an extensible location

ontology. In GI-days, pages 261–276.

NTT-DoCoMo (2007). i-mode, http://www. nttdocomo.

co.jp/english/service/imode/index.html.

Sashima, A., I. N. K. K. (2004). Consorts: A multiagent ar-

chitecture for service coordination in ubiquitous com-

puting. In Multi-Agent for Mass User Support, Post-

Proceedings of MAMUS-03, Lecture Notes in Artiﬁ-

cial Intelligence 3012, volume 3012, pages 39–46.

Sonntag, D. (2005). Towards interaction ontologies for mo-

bile devices accessing the semantic web. In 4th Inter-

national Workshop on HCI in Mobile Guides.

WSMO (2005). Web Service Modeling Ontology,

http://www.wsmo.org/.

ICEIS 2007 - International Conference on Enterprise Information Systems

190