architecture for multidomain information integration · architecture for multidomain information...
TRANSCRIPT
![Page 1: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/1.jpg)
AgentLink TFG MeetingJune 30th, 2004
Silvana Aciar, Gustavo González, Beatriz López, Josefina López-Herrera, Josep Lluís De la Rosa
http://eia.udg.es/arl/
Agents Research Lab
Universitat de Girona
Architecture for multidomain information integration
![Page 2: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/2.jpg)
2
Outline
3. Introduction
4. Our Approach: MasUM: SUM + MasID
5. MasID Methodology6. MasID Architecture
7. Preliminary results8. Conclusions and open issues
![Page 3: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/3.jpg)
3
1. Introduction
Motivation
– IRES Project: On the Integration Restaurant Services ACNET.02.50
Content-based filtering through CBR engineOpinion-based filtering through trust
Distributed multi-agent recommender system of service agents and personal agents
Available at: http://arlab.udg.es/
Special Prize to the best system deployed in the AgentCities network
![Page 4: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/4.jpg)
4
1. Introduction
Personalisation Servicesfor Open Environments
in Multiple Domains
Non-intrusive user models
Cold start problem
Domain ADomain A Domain BDomain B
DistributedDistributedHeterogeneousHeterogeneous
InformationInformationSourcesSources
SuppliersSuppliers
Collaborative-basedCollaborative-basedUser ModelsUser Models
Content-based Content-based User ModelsUser ModelsContent-based Content-based
User ModelsUser ModelsCollaborative-basedCollaborative-based
User ModelsUser Models
Multi-Agent based User Models
for Multiple Services (MasUM)
![Page 5: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/5.jpg)
5
2. Our approach
Multi-agent based User Models for Multiple Services (MasUM)
DOMAIN 1
User
Service
INTA
QUA
DOMAIN 2
Pin
Serv
ices
Attri
bute
s
User
s
Pin
Service
INTA
QUA
Pin
Se
rvic
es
Attri
bu
tes
Use
rs
Pin
SUM: Smart User Model
MasID: Multi-Agent System Integrator of
Different Domains
![Page 6: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/6.jpg)
6
Distributed Heterogeneous Information Integration
Information selection about users and services.
Heterogeneous and distributed information sources in several domains.
Information access can be established by means of relationships between the user and services.
Recomendation
Domain 1 Domain 2 Domain 3
Information sources
Unified format –Data Model
Information integrationSimilars Attributes
User database
Services database
User database
Services database
User database
Services database
Service
Attributes
User Service
Attributes
User Service
Attributes
User
3. MasID Methodology
![Page 7: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/7.jpg)
7
Information
Transactions
Obtaining characterization of products
Applying quality measure
PRODUCT, attribute, optimal weigth
Applying cluster algorithm
Saving stereotypes
Domain 1
USER, attribute, weigth
Information
Transactions
Domain 2
Integrating information-Applying similarity measure at the attributes
Making recommendation
3. MasID Methodology
Saving stereotypes
Step 1
Step 2
Step 3
Step 4
![Page 8: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/8.jpg)
8
3. MasID Methodology. Step 1
0.850.751CAN PERET1054663018421
0.900.500.75XINES GRAN MURALLA1054663018421
0.950.750.50PIZZER-A ADRIANO1054663018421
Originality……AmbientQuality-ProceRestaurantUser
0.850.751CAN PERET1032969360861
0.900.500.75XINES GRAN MURALLA1032969360861
0.500.921PIZZER-A ADRIANO1032969360861
Originality……AmbientQuality-Proce RestaurantUser
0.850.751CAN PERET1033550555498
0.900.500.75XINES GRAN MURALLA1033550555498
0.900.801PIZZER-A ADRIANO1033550555498
Originality……AmbientQuality-Proce RestaurantUser
0.750.751CAN PERET0.850.540.60XINES GRAN MURALLA0.900.801PIZZER-A ADRIANOOriginality……AmbientQuality-Proce Restaurant
![Page 9: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/9.jpg)
9
Where Pij = weight given to the attribute j of by user i
3. MasID Methodology. Step 1
Qij=1−∑i=1
k
∥Pij−POj∥⋅log2∥Pij−POj∥
POj=¿ P je
¿
P j =¿∑i=1
k
Pijk
¿
![Page 10: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/10.jpg)
10
3. MasID Methodology. Step 2
0.850.751CAN PERET1054663018421
0.900.500.75XINES GRAN MURALLA1054663018421
0.950.750.50PIZZER-A ADRIANO1054663018421
Originality……AmbientQuality-ProceRestaurantUser
0.850.751CAN PERET1032969360861
0.900.500.75XINES GRAN MURALLA1032969360861
0.500.921PIZZER-A ADRIANO1032969360861
Originality……AmbientQuality-Proce RestaurantUser
0.850.751CAN PERET1033550555498
0.900.500.75XINES GRAN MURALLA1033550555498
0.900.801PIZZER-A ADRIANO1033550555498
Originality……AmbientQuality-Proce RestaurantUser
0.900.500.751054663018421
0.850.7511054663018421
0.900.8011033550555498
Originality……AmbientQuality-ProceUser
![Page 11: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/11.jpg)
11
Where Pij = weight given to the attribute j of by user i
3. MasID Methodology. Step 2
Qij=1−∑i=1
k
∥Pij−P j∥⋅log2∥Pij−P j∥
P j =¿∑i=1
k
Pijk
¿
![Page 12: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/12.jpg)
12
3. MasID Methodology. Step 3
0.900.500.751054663018421
0.850.7511054663018421
0.900.8011033550555498
Originality……AmbientQuality-ProceUser
Cluster of users stereotypes
![Page 13: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/13.jpg)
13
3. MasID Methodology. Step 4
Open issueDepends on:
The user is in both domainsStereotypesSimilarity measuresCorrelation measures….
![Page 14: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/14.jpg)
14
4. MasID Architecture
The INTegrator Agent (INTA): Access to the data set in each domain.Computes step 1 and 2. Keeping update the relation in each domain.
QUantifier Agent(QUA):
Apply the cluster algorithm (step 3)
RECommender Agent (RECA):Computes similarities (step 4)
![Page 15: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/15.jpg)
15
In the restaurant domain the product was characterized by the optimal weight allocated to each attribute (step 1)
4. Preliminary results
Figure 1.1 Results Restaurant Domain
![Page 16: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/16.jpg)
16
5. Preliminary results
In the insurance domain the product was characterized by the weight allocated to each attribute by an expert (step 2). (no step 1 is performed)
Table 1.1 Results Insurances Domain
![Page 17: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/17.jpg)
17
6. Conclusions and future work
Multi-domain integration information – Methodology
– Architecture
– Preliminary results on off-line transactions and recommender system
Complete methodology– Open issues regarding similarities across domain
Future work– Test the integration architecture to gather user information from web services.
– Test the integration architecture to fill up the features of the Smart User Model in e-business applications
– Extend this approach to Grid environment in order to create and manage complex knowledge discovery applications composed as workflows that integrate data sets, provided as distributed services on a Grid.
![Page 18: Architecture for multidomain information integration · Architecture for multidomain information integration. 2 Outline 3. Introduction 4. ... Silvana Aciar, Gustavo González, Beatriz](https://reader031.vdocuments.net/reader031/viewer/2022022710/5bfd05ca09d3f297368be504/html5/thumbnails/18.jpg)
18
THANK YOU FOR YOURATTENTION!
Silvana Aciar, Gustavo González, Beatriz López, Josefina López-
Herrera, Josep Lluís De la Rosa http://eia.udg.es/arl/