Speed dating Classification

1
Speed dating ClassificationWhat you should know
about dating

• Stephen Cohen
• Rajesh Ranganath
• Te Thamrongrattanarit

2
Speed dating

• A rabbi invented speed dating 10 years ago
• Heres how it works
• Goal To find the model that predicts men and
  womens decisions

3
Massive feature extraction

• Easy things
• Word count
• Count of certain words
• Backchannelling
• Post-conversation word count
• Question count
• Non-academic discussion
• Etc.

• Difficult things
• Latent Dirichlet Allocation
• Latent Semantic Analysis
• Various vector similarity metrics
• Speed of conversation
• Etc.

4
Classifiers and other techniques

• Lexical Feature Extraction
• Logistic Regression with linear kernel
• Support Vector Machines with
• Linear kernel
• RBF kernel

5
Evaluation

• Principle Component Analysis
• For every feature we add, we capture more
  variance. good sign
• The Rajesh Metric for evaluating models
• Logistic Regression and SVM work just as well.
• Pick the best model based on the Rajesh Metric
• Analyze regression coefficients of the best model
  

6
What you should know about dating

• Men are more likely to say yes if ..
• More positive words are uttered. lexical
  features
• Men and women talk about the same topics Latent
  Dirichlet Allocation and Jenson-Shannon
  similarity
• Menwomen word count ratio is high
• Women ask more questions! count of question
  marksbut opposite effect on women
• And more
• Womens decisions can hardly be predicted by the
  model. (Women are hard to understand)
• Women are more likely to say yes if they talk
  about the past.
• Physical appearance? Voice? Speech? Chemistry?

7
Acknowledgement

• Professor Dan Jurafsky (Linguistics Dept.)
• Professor Dan McFarland (School of Education)
• Stephan Stiller (Computer Science)
• David Hall (Symbolic Systems and CS)