Moderated Class-membership Interchange in Relational Classification - - PowerPoint PPT Presentation

Moderated Class-membership Interchange in Relational Classification

Peter Vojtek, Mária Bieliková Slovak University of Technology in Bratislava Faculty of Informatics and Information Technologies Bratislava, Slovakia

Relational (collective) classification

Sort new scientific articles into two classes:

• articles concerning Hardware
• articles concerning Software

Relational (collective) classification

Sort new scientific articles into two classes:

• articles concerning Hardware
• articles concerning Software

article

text

Relational (collective) classification

Sort new scientific articles into two classes:

• articles concerning Hardware
• articles concerning Software

article

text

• keyword

author

Relational (collective) classification

Sort new scientific articles into two classes:

• articles concerning Hardware
• articles concerning Software

article

text

• keyword

author

relation: hasKeyword relation: hasAuthor

Relational (collective) classification

Sort new scientific articles into two classes:

• articles concerning Hardware
• articles concerning Software

article

text

• keyword

author

relation: hasKeyword relation: hasAuthor relation: references

Collective Inferencing

• classmembership of each instance is initialized

– Article No.1: [Hardware: 80%, Software: 20%] – Article No.2: [Hardware: 50%, Software: 50%]

Collective Inferencing

• classmembership of each instance is initialized

– Article No.1: [Hardware: 80%, Software: 20%] – Article No.2: [Hardware: 50%, Software: 50%]

• neighbouring instances share and update their
• neighbouring instances share and update their

classmemberships

• iterative, convergent process
• at the end

– Article No.1: [Hardware: 91%, Software: 9%] is assignet to class Hardware

Misclassification

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference

Misclassification

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference HW

HW 55% SW 45% HW 30%

article

HW

article article

SW 45% HW 30% SW 70%

Misclassification

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference HW

HW 55% SW 45% HW 30%

article

HW

article article

SW 45% HW 30% SW 70%

article

HW

article article

HW 30% SW 70% HW 40% SW 60%

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference

article

HW

article article

HW 55% SW 45% HW 30% SW 70%

article

HW

article article

HW 30% SW 70% HW 40% SW 60%

article

HW

article article

HW 40% SW 60% HW 70% SW 30%

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference

article

HW

article article

HW 55% SW 45% HW 30% SW 70%

article

HW

article article

HW 30% SW 70% HW 40% SW 60%

article

HW

article article

HW 40% SW 60% HW 70% SW 30%

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference

article

HW

article article

HW 55% SW 45% HW 30% SW 70%

article

HW

article article

HW 30% SW 70% HW 40% SW 60%

article

HW

article article

HW 40% SW 60% HW 70% SW 30%

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference

article

HW

article article

HW 55% SW 45% HW 30% SW 70%

article

HW

article article

HW 30% SW 70% HW 40% SW 60%

article

HW

article article

HW 40% SW 60% HW 70% SW 30%

?

real class

article

HW initial classmembership

article

HW 70% SW 30%

final classmembership

article

HW 80% SW 20%

Naïve Bayes collective inference

article

HW

article article

HW 55% SW 45% HW 30% SW 70%

article

HW

article article

HW 30% SW 70% HW 40% SW 60%

article

HW

article article

HW 40% SW 60% HW 70% SW 30%

?

Information Exchange Moderation

• Classmembership is evaluated using entropy

– great value: [hardware: 100%, software: 0%] – worthless: [hardware: 50%, software: 50%]

HW 100% SW 0%

50 : 50 0 : 100 hardware : software accept ignore

Information Exchange Moderation

• Classmembership is evaluated using entropy

– great value: [hardware: 100%, software: 0%] – worthless: [hardware: 50%, software: 50%]

50 : 50 0 : 100 hardware : software accept ignore threshold moderation

Information Exchange Moderation

• Classmembership is evaluated using entropy

– great value: [hardware: 100%, software: 0%] – worthless: [hardware: 50%, software: 50%]

50 : 50 0 : 100 hardware : software accept ignore linear continuous moderation

Experimental Evaluation

• MAPEKUS dataset http://mapekus.fiit.stuba.sk
• two accuracy values are compared:

– accuracy after initialization (naïve Bayes based on text of article’s abstract) (naïve Bayes based on text of article’s abstract) – accuracy after collective inferencing

Experimental Evaluation

Initialization:

• 80% accuracy

Collective inferencing:

• 89% accuracy

• 2

Experimental Evaluation

Initialization:

• 80% accuracy

Collective inferencing:

• 89% accuracy

• 2

accept ignore

Conclusions

• new method to increase robustness of

relational classification

• succesfully evaluated on

– MAPEKUS dataset – MAPEKUS dataset – (Netflix+IMDB) movie recommendation

• further work:

– different shapes of moderation function – shift to homophily