Paper: Learning Combination Features with L1 Regularization

ACL ID N09-2025
Title Learning Combination Features with L1 Regularization
Venue Human Language Technologies
Session Short Paper
Year 2009
Authors
  • Daisuke Okanohara (University of Tokyo, Tokyo Japan)
  • Jun'ichi Tsujii (University of Tokyo, Tokyo Japan; University of Manchester, Manchester UK; National Center for Text Mining, UK)

When linear classifiers cannot successfully classify data, we often add combination fea- tures, which are products of several original features. The searching for effective combi- nation features, namely feature engineering, requires domain-specific knowledge and hard work. We present herein an efficient algorithm for learning an L1 regularized logistic regres- sion model with combination features. We propose to use the grafting algorithm with ef- ficient computation of gradients. This enables us to find optimal weights efficiently without enumerating all combination features. By us- ing L1 regularization, the result we obtain is very compact and achieves very efficient in- ference. In experiments with NLP tasks, we show that the proposed method can extract ef- fective combination features...