BayesANIL - A Bayesian Model for Handling Approximate, Noisy or Incomplete Labeling in Text Classification

Posted in Science on September 08, 2008


BayesANIL - A Bayesian Model for Handling Approximate, Noisy or Incomplete Labeling in Text Classification

Lecture slides:

  • BayesANIL A Bayesian Model for Handling Approximate, Noisy or Incomplete Labeling in Text Classification
  • Outline
  • Motivation - hurdles in supervised learning of text classifiers
  • Related work
  • Related work (contd.)
  • What we propose
  • Role of BayesANIL in text classification
  • The BayesANIL model : notations
  • The BayesANIL model: Objective function
  • The BayesANIL model: E and M Steps
  • The Algorithm
  • Re-estimating the empirical distribution
  • Utilizing parameters of BayesANIL in NB
  • Utilizing parameters of BayesANIL in SVM
  • Experiments and Results
  • Experiments and Results: Supervised
  • Experiments and Results: Labeled-unlabeled
  • Experiments and Results: Access to unlabeled for WebKB
  • Experiments and Results: Access to unlabeled for 20 Newsgroups
  • Experiments and Results: Noisy Labels
  • Comparison with results as reported by (Bing Liu et al 2003)
  • Experiments and Results: Notion of Support
  • Summary
  • Future work

Author: Ganes Ramakrishnan, Ibm

Watch Video

Tags: Science, Lectures, Computer Science, Machine Learning, VideoLectures.Net, Bayesian Learning