technical session i mcrdr classifier & database design presenter yangsok kim school of computing

25
Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

Upload: miranda-parsons

Post on 17-Jan-2016

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

Technical Session I

MCRDR Classifier & Database Design

Presenter

Yangsok Kim

Sch

ool

of C

om

put

ing

Page 2: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Scope

• Explain MCRDR classifier installation process• Explain MCRDR classifier database table

structure• Explain relationship between classification

process and database table structure

Page 3: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

• Download MCRDR classifier from the Unit Web page (http://www.comp.utas.edu.au/iWeb/iPrj/kxa459/2007/project.html)

• Create a folder for MCRDR (e.g., mcrdr_classifier) and unzip the download file

• Execute MonClassifier_KXA4592007.exe• Configure database setup

MCRDR classifier InstallationClick this icon

Select “Database”

Fill in database information

(host, userid, password, database name)

Click reconnect

button

Select username and fill in username and passowrd

Page 4: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Database

• You can see your database via phpMyAdmin interface (http://www.comp.utas.edu.au/iWeb/phpMyAdmin/)

• 9 tables are client side (classifier table) and 7 tables are server side table.

Server Tables

Client (Classifier) Tables

Page 5: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Database – Server Side Tablestbmonservsite

tbmonservsitetree

tbmonservarticle

Join tbmonservsite and tbmonservsitetree

SELECT a.MonSiteID, b.MonSitePID, a.MonSiteURL,

b.MonSiteName, b,nTime FROM tbmonservsite a,

tbmonservsitetree b WHERE a.MonSiteID=b.MonSiteID

Page 6: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Import Monitoring WebpagesClick this icon to start

import web sites

Select Checkbox to import monitoring web sites

Page 7: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Import Monitoring Webpagestbclassify_user_article

tbmonservarticle

If you want to article details, you need to join tbclassify_USERNAME_article and tbmonservarticle

e.g.SELECT b.ArticleID, b.Title, b.Body FROMtbclassify_user_article a, tbmonservarticle b WHERE a.ArticleID=b.ArticleID

This table date is filled when you import

monitoring Web pages.

OR

After registering Web pages, the MonServer

automatically feeds newly collected articles.

Page 8: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Import Monitoring Webpages

TypeID illustrates current status of Article0:Not Classified1:Classified2:Deleted

ClassifiedDate This time information is saved when the article classified

This field illustrates whether the article is read or not

tbclassify_user_article

Page 9: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Import Monitoring Webpagestbclassify_user_monsitekwdtree

Monitoring Site ID

Monitoring site name

tbmonservsite tbmonservsitetree

Page 10: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Classifier Main UIMonitoring Site Listtbclassify_user_monsitekwdtree

Monitored Article Listtbclassify_USERNAME_article and

tbmonservarticle

Selected Articletbclassify_user_article and

tbmonservarticle

Classification structuretbclassify_user_foldertree

Inference Resultstbclassify_user_articlefolder,

tbclassify_user_ruletreetbclassify_user_rulecondition

Page 11: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule base

• MCRDR classifier maintains rule base for classification

• Rule base is an n-ary tree• Each node is a rule• Rule consists of condition parts and conclusion

– Condition Attributes are Title, Body, and All of a article– Each condition evaluates whether the condition words

is included in the articles– Conclusion is a folder of the classification structure

Page 12: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Base

IF

THEN

Root

A

B

C

Title: a,b,c,g,f

Body: c, d, k,m

ClassificationStructure

Article

Root

Rule 1

Rule 2

Rule 3

Rule 4

Title includes {a, b}AND

Body include {c, d}

Classify into folder C

Rule Base

Page 13: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Base tbclassify_user_ruletree

tbclassify_user_rulecondition

Each rule has parent rule idThere are two types of rules(refine

rule(0), stop rule(1)Conclusion ID is a folder ID

CornerstoneCaseID is an ArticleID that is used for rule creation

Each rulehas multiple conditions

Condition word, which is case sensitiveThree condition types – 0:Title,

1:Body, and 2:ALL

Page 14: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule base

tbclassify_user_articlefolder

This table stores classification

results

Page 15: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Classification Structure

• There are predefined classification (folder) structure. Don’t change any folders!!

• Each folder is used as the conclusion of rule• Classification structure is a tree.

Page 16: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Classification Structure

tbclassify_user_foldertree

Each folder’s parent folder

ID

Each folder has unique ID

Page 17: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Creation

• With MCRDR classifier, user incrementally creates rules that classify documents.

• Rule creation initiated – 1) when the classifier doesn’t provide

recommendation or – 2)when the classifier gives wrong recommendation

• Rule can be refine rule or stop rule

Page 18: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Creation - BasicSelect Article

form the Article List

Click Knowledge Acquisition icon

Select a folder that this article is

classified

Select condition word

Select condition type

Condition word is added here

Display articles that satisfies current rule.

We need to validate these

casesSelect cases that should

not classified by current rule

System place the selected article in this

pane

System shows keywords that are not found in the

excluded cases.

Click this button to create a rule

New rule added to the rule tree

If you want to finish click this

button

Page 19: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Creation - Basictbclassify_user_ruletree

tbclassify_user_rulecondition

tbclassify_user_articlefolder

Page 20: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Refine Rule

View current recommendation

Select article that should be refinedClick knowledge

acquisition icon

Select a folder under the current

recommended folder

The following process is same the basic rule creation

process

New rule is added under the parent rule

Page 21: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Creation – Refine Ruletbclassify_user_ruletree

tbclassify_user_rulecondition

tbclassify_user_articlefolder

New Rule is added

Classification result has been changed

Page 22: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Creation – Stop ruleSelect an article

that should not be classified this

folder

Select current recommended folder and choose deselect

menu from the shortcut menu

After selecting “Deselect” menu,

follow the basic rule creation process

Stop rule is added under the current fired rule

Page 23: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Rule Creation – Stop rule

tbclassify_user_articlefolder tbclassify_user_articlefolder

This conditions are added

This result is deleted

This rule is added. Note the RuleType and ConclusionID

Page 24: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Inference

Page 25: Technical Session I MCRDR Classifier & Database Design Presenter Yangsok Kim School of Computing

School of Computing,University of Tasmania

Q&A