MachineLearningfor Language(Technology( Lecture5...

17
Machine Learning for Language Technology Lecture 5: Bayesian Classifica3on Marina San3ni Department of Linguis3cs and Philology Uppsala University, Uppsala, Sweden Autumn 2014 Acknowledgement: Thanks to Prof. Joakim Nivre for course design and materials

Transcript of MachineLearningfor Language(Technology( Lecture5...

Page 1: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Machine  Learning  for  Language  Technology    Lecture  5:  Bayesian  Classifica3on  

Marina  San3ni  Department  of  Linguis3cs  and  Philology  Uppsala  University,  Uppsala,  Sweden  

 Autumn  2014  

 Acknowledgement:  Thanks  to  Prof.  Joakim  Nivre  for  course  design  and  materials  

Page 2: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Bayesian  Classifica3on  (i)  

Page 3: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Bayesian  Classifica3on  (ii)  

Page 4: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Instance  APributes  

Page 5: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Ex:  Spam  Filtering  

Page 6: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Naive  Bayes  Classifiers  

Page 7: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Modeling  

Page 8: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Inference  

Page 9: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Learning  (i)  

Page 10: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Learning  (ii)  

Page 11: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Learning  (iii)  

Page 12: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Evalua3on  (i)  

Page 13: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Evalua3on  (ii)  

Page 14: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Mul3-­‐Class  Classifica3on  

Page 15: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Naive  Bayes  in  NLP  

Page 16: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

Tips  and  Tricks  

Page 17: MachineLearningfor Language(Technology( Lecture5 ...santini.se/teaching/ml/2014/Lecture5_Bayesian_Classification.pdf · Word sense disambiguation Key properties: Attributes can be

The  end