》》Balancing between over-weighting and under-weighting in supervised term weighting

Start: 2017/10/30
Finish: 2017/10/??

PDF
Slide not available yet

Very Similar Papers on context of supervised term weight

• 2015, New term weighting schemes with combination of multiple classifier for sentiment analysis PDF READ
• 2007, Supervised and Traditional Term Weighting
  Methods for Automatic Text Categorization PDF
• 2005, Learn to Weight Terms in Information Retrieval Using Category Information PDF READ
• 2009, Learning term-weighting functions for similarity measures PDF

IDEAs of Application

• some supervised global term weighting can be useful to score term when genre-filtering

ABSTRACT

• Supervised term weighting for text categorization.
• Proposed a new concept of Over-weighting, in context of controlling over-weighting and under-weighting
  • over-weighting: assign larger weight to terms with more imbalanced distributions across categories
  • under-weighting: ratios between term weights becomes too small. Because sublinear scaling and bias term shrink the ratios
• Proposed a new supervised term weighting scheme, regularized entropy (re)
• Present 3 regularization techniques: add-on smoothing, sublinear scaling, bias term
• Evaluated on a lots of datasets on both tasks of topical classification and sentiment classification
• Other terminology:
  • regularization and over-weighting & under-weighting
  • scaling functions
  • bias term

INTRODUCTION

REVIEW of TERM WEIGHTING SCHEMES

| weight(term, doc) = local_weight_of_term x global_weight_of_term x normalization_factor_of_doc |

• Local term weighting

• Global term weighting

...