User Care Preference-based Service Discovery in a Ubiquitous Environments Dongpil Kwak, Joongsoo Lee, Dohyun Kim, and Younghee Lee Talk by Joongsoo Lee.

User Care Preference-based Service Discovery in a Ubiquitous Environments

Dongpil Kwak, Joongsoo Lee, Dohyun Kim, and Younghee Lee

Talk by Joongsoo Lee

[email protected]

Information and Communications Univ,

Daejeon, Korea

2

Introduction

• Ubiquitous or Pervasive Computing Environments

• Networking everywhere

• Providing convenient environment according to human’s context while

minimizing human’s intervention

• Semantic service discovery, context management, and inference engine

are important building blocks

• Service Discovery in Ubiquitous Environments

• Service requester is human or a proxy device

• It can be occurred in background

• Most appropriate service should be returned or the services should be

ranked based on user’s context

3

Introduction

• Context and user preference

• As an example, messaging application

In front of a laptop

Residing in “Room B”

Attends a meeting

Prefers visible interface

Prefers bigger screen

User’s Context Preference

Big screen in Room B

Laptop

Cell phone

Speaker in Room B

Laptop

Service discoveredCandidate services

Servicefiltering

4

Problem Identification

• Questions in mind

• How to set up user preference?

• How to valuate user preference?

• How to retrieve appropriate service

based on user preference?

• We are focusing on user care applications

• Such as healthcare services

• Using the five senses (sight, hearing, taste, smell, touch)

Prefers visible interface

Prefers bigger screen

Preference

5

Wearable Sensors in Future

A figure from NTT Docomo [Hirotaka ’03]

6

Overall Architecture

• A user experiencing specific symptom make a service discovery

request to know sensitive services

or

Trigger

7

Service Discovery Procedure

Service matching with service ontology

User care preference * service effect

Preprocessing knowledge set

Service ranking & retrieval

Service Query Alarming

TV, FM radio, clock, light

Hearing effect -Sight effect +

Light, TV

8

User care preference policy

• Preference setting• Using medical database and IR techniques• Multiple effects

• To match factor of service

• Ex) highly quiet preferred and leave in peace sense of sound (-), sense of sight and sound (-)

• Weight values

• To measure the degree of relativity with functional effects of service

• Ex) highly quiet preferred and avoid blue the term ‘highly’ indicating weight

• Each capacity of effect is independent one another

• -1 <= each sense of user care preference <= 1

9

User care preference policy

10

Service Ontology based on Functional Effect

11

Service Matching

• User care preference (Hearing, Sight)=(0.7, 0.5)• Consider uncertainty of weighted value (Smoothing with Error rate)

• Ex) 15% error rate => (0.7, 0.5, 0.18) , 0% => weighted sum applied

• Normalization User care preference (hearing, sight, uncertain)=(0.51, 0.36, 0.13)

• Service (Hearing, Sight, Touch)=(0.6, 0.3, 0.4)• Each value is the degree of satisfaction among elements of evidence

12

Service Matching

Sound (0.51) Sight (0.36) Uncertain (0.13)

Sound (0.6) Sound (0.31) Conflict(0.21) Sound (0.08)

Sight (0.3) Conflict(0.15) Sight (0.11) Sight (0.04)

Touch (0.4) Conflict(0.2) Conflict (0.15) Touch (0.05)Basic probability assignment (negativeSound)= -(0.31+0.08) = -

0.39 Basic probability assignment (negativeSight)= -(0.11 + 0.04) = -

0.15 Basic probability assignment (touch)=0.05

UserCarePreference(Headache)= UserCarePreference(negativeSound, negativeSight, Uncertain)= UserCarePreference(-(0.51), -(0.36), 0.13)

Service(CellPhone)= Service(Sound, Sight, Touch)= Service(0.6, 0.3, 0.4)

Relativity(CellPhone)= Relativity(negativeSound, negativeSight, Tough)= Relativity(-(0.39), -(0.15), 0.05)= -0.49

13

Implementation

• Mobile object

• Location recognition and symptom perception

• Medical symptom analyzer in context manager

• Symptom analysis

• Service matching module in service manager

• Measurement of relative degree with services

14

Symptom analysis

• Medical symptom follows

predefined rules to increase

precision

• Three rules are applied for

symptom analysis

• Key terms such as ‘cold’ and ‘hot’

are picked up to compare

‘sense=touch state =negative’ and

‘sense=touch state=positive’ in

keyword table

• It picks up negative terms that

cause a reserve of meaning such as

‘not’, ‘avoid’ et al.

• It picks up terms that means weight

such as ‘highly’, ‘relatively’ et al.

15

Service matching design

• Functional effects of service are

described in abstraction by

ontology

• Functional effects are classified by

five senses

• Its functional effect consists of

service action

• Service action consists of services

16

Summary

• Conclusion

• This work deals with situation when service discovery is executed

according to user’s condition

• Classified service query

• Design for matching with classified query and service

• Expected to increase the satisfaction of users

• Make richer to represent user’s requirement

• Increase service matching

• Applied to Healthcare, User Preference based Service Discovery

• Future work

• More study on human sense & modeling