AgentSeeker - AgentService

wrackbaaΚινητά – Ασύρματες Τεχνολογίες

10 Δεκ 2013 (πριν από 3 χρόνια και 8 μήνες)

69 εμφανίσεις

AgentSeeker:

an Ontology
-
based Enterprise Search Engine

Università degli Studi di Genova

Facoltà di Ingegneria Informatica

l . i . d . o .



DIST

Torino, 10th September 2009

Andrea Passadore, Alberto Grosso

and Antonio Boccalatte

AgentSeeker


MALLOW AWESOME 009

2

What is AgentSeeker?

Personal Hard
-
disk

Local intranet

Internet

Ontology Support

End Users

Multi
-
Agent Platform

What is AgentSeeker?


AgentSeeker

is a search engine aimed to
enterprise applications because it is:


Scalable


Robust


Flexible


Open to third
-
party programs

AgentSeeker


MALLOW AWESOME 009

3

Additional Platform

Third
-
party applications


Developers:


The

l.i.d.o.

staff of the University of
Genova


The project is…


…based on the .NET framework 3.5


Alternatively, the mono project


…written completely in C#


…distributed under the LGPL license


…available on
www.agentservice.it



AgentSeeker


MALLOW AWESOME 009

4

AgentService

AgentSeeker


MALLOW AWESOME 009

5

Core systems

Agent model

Services

Behaviours

Knowledges

Tools

Storage

Messaging

Persistence

Logging

AMS

DF

MTS

Ontologies

Monitoring

Roles code

generator

Maintenance

Development

Design tools

APX

Agent mobility

Wizards

Platform

infrastructure

AgentService

AgentSeeker


MALLOW AWESOME 009

6

AMS

DF

MTS

Core Modules

storage

messaging

persistence

logging

Additional Modules


Agent Management System

-

agents management

-

scheduling of agent activities

-

platform services supervision

-

white pages services


Directory Facilitator

-

agents localization

-

yellow pages services


Message Transport System

-

agent communication

-

message dispatching

FIPA Service Components (Agents)

Behaviours

Knowledge objects

The AgentService ecosystem

AgentSeeker


MALLOW AWESOME 009

7

Internet

External Application

AgentService Federation


Why federated platforms?


To delocalize services


To balance the computational load


What does it mean to federate platforms?


To dispatch messages to remote agents


To coordinate the yellow pages services


To move agent instances among platforms



In a totally transparent way for the developers
and end
-
users.

AgentSeeker


MALLOW AWESOME 009

8

The AgentSeeker platform

AgentSeeker


MALLOW AWESOME 009

9

Indexing agents

Manager

Merger

Yellow Pages Service

Ontology agent

Query agent

Admin console pseudo
-
agent

Web page pseudo
-
agent

User

Administrator

AgentSeeker Federation

AgentSeeker


MALLOW AWESOME 009

10

User

Load balancing

AgentSeeker


MALLOW AWESOME 009

11

A free indexer?

There is Agent#2

Index agents009.di.unito.it

Remove me

A free indexer?

There is Agent#1

Index awesome009.disi.unige.it

Remove me

Subscribe me

Subscribe me

AgentService Federation Management Suite


It manages a cloud of
AgentService

platforms…


… adding new nodes to the cloud


… monitoring the availability of connected nodes.


… managing the whole platform life
-
cycle


… managing the agent instances:


Creation


Start


Stop


Movement


… managing agent applications.


AgentSeeker


MALLOW AWESOME 009

12

AgentService Federation Management Suite

AgentSeeker


MALLOW AWESOME 009

13

Administrator

Web Service

Node Discovery

Performances monitoring

AgentService Federation Management Suite


The performance monitoring is based on:


CPU usage


Free RAM


Number of running behaviours



Every node…


… performs a node discovery process


… cooperates with peers to share the topology.

AgentSeeker


MALLOW AWESOME 009

14

AgentService Federation Management Suite

AgentSeeker


WOA 2009

15

The ontology agent


The ontology agent…


... supports owl


rdf ontologies


... maintains ontologies


From the official company knowledge base


Uploaded by end users

AgentSeeker


MALLOW AWESOME 009

16

A priori classification

Classification of results

Query expansion

Classifying documents

17

Clean

stop
words

Word
Stemming

Count

occurrences

mechanics
.owl: 0.32

automotive
.owl: 0.09

energy
.owl: 0.0002

AgentSeeker


MALLOW AWESOME 009

Classifying documents


The
a priori
classification…


… is useful to first privilege the on topic
documents in the results of a user query.


… is useful to descrease the number of false
positive hits.


Running during the index sessions…


Supports only the ontologies currently in the
repository.

18

AgentSeeker


MALLOW AWESOME 009

Clustering results

19

Automotive.owl

End User

AgentSeeker


MALLOW AWESOME 009

car

coupe

Station wagon

convertible

“Costs”

motor

Electric

motor

Costs AND car

Costs AND coupe

Costs AND convertible

Costs AND station wagon

Costs AND motor

Costs AND electric motor

car

coupe

convertible

station wagon

Clustering results


The ontological clustering…


… is useful to self organize the query results…


… in folders reflecting the conceptual organization of
an argument


… is useful to orientate oneself into a relatively
new argument.


… can be based on imported ontologies



… requires more resources and process time.

20

AgentSeeker


MALLOW AWESOME 009

Query expansion

21

Automotive.owl

Extend by:

subclasses

properties

synonyms

motor

piston

valve

engine

“Ferrari motor”

“Ferrari motor” OR

“Ferrari piston” OR

“Ferrari valve”

End User

AgentSeeker


MALLOW AWESOME 009

Query expansion


The query expansion: +250% of found
documents


in the first 60 documents:


11% of documents are off topic (with query
expansion)


35% of documents are off topic (without query
expansion)

22

AgentSeeker


MALLOW AWESOME 009

Some evaluations


To build an ontology for supporting users to
find documents…


… is a complex task


… is a time
-
consuming job


… requires a lot of enterprise resources.



Solution:


To develop a sort of wikipedia for constructing
ontologies.

23

AgentSeeker


MALLOW AWESOME 009

Ontological wiki

24


Add concepts and
relations


Modify concepts


Track changes


Evaluate user’s activities

Add

concept

Modify

concept

AgentSeeker


MALLOW AWESOME 009

An AgentSeeker deployment

25

Intel XEON

dual

core

2
GHz

1,5 GB
of

RAM

AMD
Athlon


2 GHz

960 MB of RAM

Intel Pentium 4

2 GHz

512 MB of RAM


Total: 304.000 documents


Rate: 6000 documents per hour


1000 web sites visited.


AgentSeeker


MALLOW AWESOME 009

Conclusions


AgentSeeker…


… is a multi
-
agent society


Scalable


Flexible


Robust


Open


… supports ontologies in order to model…


… the company business domains.


… the skills of the users.


26

AgentSeeker


MALLOW AWESOME 009

Future works


Ontological

wiki


To

create a social
collaboration

for

building
semantic

models


To

improve the usability of the system in term
of user’s interactions.


To improve the ontological support


By
introducing

reasoners


By

interacting

with



WordNet


Swoogle



27

AgentSeeker


MALLOW AWESOME 009

The end

28

Thank

you

for

your

attention
!

AgentSeeker


MALLOW AWESOME 009