ºÝºÝߣ

ºÝºÝߣShare a Scribd company logo

KDD Cup 2013 Track 1 - Author - Paper Identification challenge (2nd place team)

•Download as PPTX, PDF•
•
Dmitry Efimov
Dmitry Efimov

We describe our approach for solution of Author - Paper Identification Challenge: https://www.kaggle.com/c/kdd-cup-2013-author-paper-identification-challenge

1 of 12
Downloaded 22 times
KDD Cup 2013 Author – Paper Identification Challenge (2nd place team) Dmitry Efimov Lucas Silva Benjamin Solecki
Approach summary Goal: find incorrectly assigned pairs author-paper Supervised machine learning problem with binary response Deep feature engineering (> 300 features) Gradient Boosting Machine (package gbm in R)
Author – Paper graph
Author features count journals tf-idf measure Count features NLP features Multiple source features author’s duplicates
Paper features Count features NLP features Multiple source features Additional features count keywords tf-idf measure paper’s duplicates reverse features engineering
Author – paper features (1 of 4) Count features Multiple source features Additional features Likelihood features
Author – paper features (2 of 4) Count features Additional features count of coauthors with the same affiliation reverse feature engineering: year ranking feature
Author – paper features (3 of 4) Multiple source features how many times pair author-paper appeared in the Microsoft database?
Author – paper features (4 of 4) Likelihood features use Lj and Lja as features 1) use (α∙ Lj + (1−α)∙ Lja) as feature (shrunken likelihood); 2) mixed-effects models (package lme4 in R) to find α Lj – likelihood by journal Lja – likelihood by journal and author
Model Gradient Boosting Machine (package gbm in R) Grid search for the set of parameters 83 features in the final model (out of 300 calculated features )
Result and conclusion • Our MAP score is 0.98144 (the winning submission score is 0.98259). • Many algorithms (LambdaRank, LambdaMART, RankBoost) based on MAP optimization gave less MAP score than GBM with Bernoulli distribution. • The idea of feature classification based on bipartite author-paper graph is very promising. Analyzing of graph topology can give ideas for new features.
Thank you!

More Related Content

KDD Cup 2013 Track 1 - Author - Paper Identification challenge (2nd place team)

  • 1. KDD Cup 2013 Author – Paper Identification Challenge (2nd place team) Dmitry Efimov Lucas Silva Benjamin Solecki
  • 2. Approach summary Goal: find incorrectly assigned pairs author-paper Supervised machine learning problem with binary response Deep feature engineering (> 300 features) Gradient Boosting Machine (package gbm in R)
  • 4. Author features count journals tf-idf measure Count features NLP features Multiple source features author’s duplicates
  • 5. Paper features Count features NLP features Multiple source features Additional features count keywords tf-idf measure paper’s duplicates reverse features engineering
  • 6. Author – paper features (1 of 4) Count features Multiple source features Additional features Likelihood features
  • 7. Author – paper features (2 of 4) Count features Additional features count of coauthors with the same affiliation reverse feature engineering: year ranking feature
  • 8. Author – paper features (3 of 4) Multiple source features how many times pair author-paper appeared in the Microsoft database?
  • 9. Author – paper features (4 of 4) Likelihood features use Lj and Lja as features 1) use (α∙ Lj + (1−α)∙ Lja) as feature (shrunken likelihood); 2) mixed-effects models (package lme4 in R) to find α Lj – likelihood by journal Lja – likelihood by journal and author
  • 10. Model Gradient Boosting Machine (package gbm in R) Grid search for the set of parameters 83 features in the final model (out of 300 calculated features )
  • 11. Result and conclusion • Our MAP score is 0.98144 (the winning submission score is 0.98259). • Many algorithms (LambdaRank, LambdaMART, RankBoost) based on MAP optimization gave less MAP score than GBM with Bernoulli distribution. • The idea of feature classification based on bipartite author-paper graph is very promising. Analyzing of graph topology can give ideas for new features.