Higher Order Knowledge Management Platform for the Personalized ...

yawnknotManagement

Nov 6, 2013 (4 years and 3 days ago)

69 views


SITUATION
-
AWARENESS MODEL FOR HIGHER

ORDER NETWORK KNOWLEDGE MANAGEMENT
PLATFORM




Aekyung

Moon, ETRI

akmoon@etri.re.kr

Contents


Introduction


Knowledge Management Platform


Knowledge Layer


Knowledge Modeling


Situation
-
Awareness Model


Evaluations


Conclusions


2

Introduction


Service paradigm

of Telecommunications


Open & Convergence (2000s)


Open API (e.g. Parlay/OSA, Parlay X) and Open Service Gateway


Personalization and Contextualization (Now)



need Something New to support context
-
aware personalized service for next generation networks.


What we need


Allow to build personalized and adaptive communication services


Need to add the special capabilities for providing context
-
aware personalized
service on the existing network


Network Knowledge Layer


Network Knowledge Layer should be overlaid on the existing network without
any changes

What we did for Network Knowledge Layer (Context
-
aware
Personalized Services)


Design Knowledge Management Platform for implementing Network Knowledge Layer


Modeling of Network Knowledge
to handled in a value
-
added Network Knowledge Layer


Ontology Model and
Learner Algorithm for
Situation
-
Awareness Model




3

Network Knowledge Layer



Position & Role of Network Knowledge Layer


Application Layer

Control Layer or

Service Layer

Transport Layer

Access Layer

Context
-
aware Personalized Service

Knowledge
L
ayer

User Device

Context
-
aware
Personalized Service

Context
-
aware Personalized Service

To provide interfaces (API)
with context
-
aware and
personalized

mobile
application programmers


To
collect and
manipulate
network knowledge


To acquire network
knowledge

be overlaid on the existing
networks

provide
the personalized
services easily

4

Service Architecture for Knowledge
Management Platform


Design Consideration of Network Knowledge Layer

Acquisition

To acquire network
knowledge from
underlying network
and user’s terminal


What are network
knowledge and where
are they?


How to acquire
network knowledge
from sources?

Manipulation

To manipulate
network knowledge


How to build network
knowledge?


How to manipulate
network knowledge?


How to adapt current
acquired knowledge
into knowledge model?

Exposure

To provide interfaces
(API) for context
-
aware mobile
application
programmers


How to provide
network knowledge?


Which knowledge
are provided for
the user?

5

Classification of Network Knowledge



Context

Terminal

Context

User

Context

PCS

Context

Environment

Context

Service

Profile

Network

Profile

Profile

Preference

Service

Preference

Network

Preference

Device

Profile

User

Profile

Device

Preference

Usage

Behavior

Rule/

Policy

High level

Knowledge

Context

Reasoner

Learner

Low level

Knowledge

User Behavior
Pattern

User Intention

Predictor

Situation

Recommender

6

Low
-
level Network Knowledge


preference





IMS HSS

(physical location,
status, PUI, PRI, ..

MLP Server
(physical location)

Device profile

Service
Repository

IMS Presence server
(activity, mood,
place
-
is)

Parlay X GW (TL, TS, Presence)

User profile (
SSid
,
age, gender, job)

Network Profile
(Access Net., BW)

Device Profile

Service
Profile

Context




Schedule

Physical
location

Terminal
status

current

BW

Network
access type

current

network
traffic

presence

Sensing

info.

User’s device
preference

User’s network
preference

User’s service
preference

PCS(Personal Communication Sphere)

Preference


Profile

Service Provider

Device Vender

Operator’s

User information

7

Low
-
level Network Knowledge Modeling



Ontology Model


10 classes, 13 Object properties, 39
datatype

properties

User



device

service

hasDevice

hasService

hasService

isAvaliabelAt

Schedule

presence

activity

hasPresence

hasGroup

location

isLocatedIn

hasDevice

network

use

hasActivity

knows

belongToPlace

group

hasSchedule

hasPreference

Preference

8

Knowledge Management Platform


system architecture

9

Knowledge Management Platform
Situation
-
Awareness Model
Context
Reasoner
Ontology
Model
User Service
Usage Model
Usage Pattern
Learner
Lower
order
Know
-
ledge
Underlying Network (e.g. IMS)
High
order
Knowledge Exposure
Context
-
aware
Personalized Service
Context
-
aware
Personalized Service
Context
-
aware
Personalized
Service
Recommender
Services
Recommender
Contents
Recommender
Higher
order
Know
-
ledge
Service
Info.
Contents
Info.
Situation
-
Awareness Ontology Model

for Context
Reasoner

10


Class hierarchy
Object properties
Implied relation
Class
Instance
Thing
User
Network
Service
Group
Preference
Device
Activity
owns
owl:inverseOf
uses
locatedIn
isAvailableAt
hasAvailableNetwork
Schedule
Situatio
n
hasPreference
isMemberOf
owl:inverseOf
owl:inverseOf
PersonIn
Meeting
Domain Specific Ontology
PersonIn
Shopping
PersonIn
Watching
TV
Waiting
ForBus
Service
Preferenc
e
Domain
Preference
Service
Preference1
Hong
hasPreferenceValue
Service
Preference2
Next Segment
High
Domain
Preference1
Location
PlaceTy
pr
hasPreference
Service
Category
Entertainment
Commerce
Information
Communica
tion
TV
Service Ontology
Music
Drama
Action
hasLocation
User Centric Core Ontology
PersonInMeeting

User Π

hasActivity.Meeting
Π

locatedIn.Office
Π

own.
DeviceInUse

User
DeviceInUse

Device Π

isStatus.ON

Device
PersonInShopping

User Π

hasActivity.Shopping
Π(

locatedIn.Arena

locatedIn.Shopping
-
area


locatedIn.Store
)

User
PersonInWaitingForBus

User Π

locatedIn.BusStation

User
PersonWatchingTV
= User Π

has Activity.TV Π

locatedIn.Room

User
N
-
ary
relations
preferenceProbability
Learner


Step 1: the context profile has to register attribute values which each
context has


If C
-
TBL for c
k

(c
k



States) doesn’t exist, create C
-
TBL for context c
k

using context profiles and initialize
C
-
TBL and reward value R.


Step 2: Input an current action ac(t) by user selection. Determine reward R
according to user behavior information. Update the C
-
TBL as following rules:


for each
c
i

in C
-
TBL[
a
i,k
(t)
][
ac
(t)
] do


C
-
TBL[
a
i,k
][ac
(t)
]

C
-
TBL[
a
i,k
(t)
][
ac
(t)
]+

R
(t)


Context Profile


States = {c
1
, ...,
c
n
}, 1


n, c
k

:
k
th

context in the State


Attributes(
c
i
) = {a

i,1
, ... a
i,k
}, 1


k and 1


i



n


Action Classes = {ac
1
, ...
ac
m
}, 1


m


Reward R= { Selection
-
r
s
, Positive Feedback
-
r
p
, Negative Feedback
-
r
n

}




11

Evaluation (1/2)


Evaluate performance of Learner


UCI Machine Learning Repository
(
http://archive.ics.uci.edu/ml/
)


177 data sets as a service to the machine learning community




12

Data

Instance

Attr
.
(Categorical)

ActionClass

Create

Approval

665

15(9)

2

Balloons

20

4(4)

2

Balance

625

4(4)

3

Iris

150

4(0)

3

Wine

178

13(0)

3

Evaluation (2/2)


We implemented two algorithms:


Learner
-
S and Learner
-
Q. Learner
-
S uses the min
-
max normalization and quantizes
continuous value into 10 fixed intervals.


Learner

Q quantize continuous value into minimal k intervals


machine learning algorithms from the
Weka

tool
-
kit : J48,
ZeroR
,
NaiveBayes


Weka
,
http://www.cs.waikato.ac.nz/ml/weka/


Weka

is a collection of machine learning algorithms for data mining tasks


Performance Matrix


the accuracy (precision)


R : the number of recommended items to a user



RP (Recommended Preference) : the number which a user actually prefers


Precision = RP / R ( %)


k
-
fold cross validation


In order to raise the confidence of experiments


The data set is divided into k subsets, and the holdout method is repeated k times




13

Simulation Results


better than other algorithms in the aspect of Create Approval.


The precision of Our approach is 86.8% at create approval.


All experiments Our approach are better than
ZeroR

with 1.5 times


Learner
-
Q is better than other algorithms in the aspect of Iris and Wine



14

0
20
40
60
80
100
120
Leaner
J48
ZeroR
Naïve Bayses
SMO
create Approval
Balance
balloon
0
20
40
60
80
100
120
Learner
-
S
Learner
-
Q
J48
ZeroR
Naïve
Bayses
SMO
iris
wine
Conclusions


For implementing this network knowledge layer,


classify network knowledge


build ontology model to represent classified network


knowledge including user profiles and user’s preference.


the system architecture of knowledge management platform


Situation awareness model


evaluated the operations using UCI machine learning


repository and
Weka

tool
-
kit


better than other algorithms in the most of experiments

15

Thank You

16