Political standing classification task

[timothywangdev]

• [x] labels?

LEFT RIGHT

• [x] what subreddits should we use?

LEFT = ['Liberal', 'progressive', 'socialism']
RIGHT = ['Conservative', 'republicans']

• [x] using 2014,2015 subreddit data

Only tested it on 2015.3 dataset

• [ ] model

  • [x] NaiveBayes

  MultinomialNB: Training accuracy: 0.765488302126 Testing accuracy: 0.688398052488
           precision    recall  f1-score   support
  
     Left       0.66      0.96      0.79      5032
    Right       0.83      0.28      0.42      3389
  
  avg / total       0.73      0.69      0.64      8421

  • [x] RandomForest (40 estimators)

  Training accuracy: 0.995685048098 Testing accuracy: 0.689348058425
             precision    recall  f1-score   support
  
     Left       0.69      0.87      0.77      5026
    Right       0.68      0.43      0.53      3395
  
  avg / total       0.69      0.69      0.67      8421

  • [x] AdaBoostClassifier (Default setting, DecisionTrees)

  Training accuracy: 0.669371758838 Testing accuracy: 0.651941574635
             precision    recall  f1-score   support
  
     Left       0.65      0.88      0.75      4977
    Right       0.65      0.32      0.43      3444
  
  avg / total       0.65      0.65      0.62      8421

• [x] evaluation (25% test data)

[timothywangdev]

A basic classifier has been added to /politicalParty 71f70ee40f286f0fee7daf5e5f8ab0bbd8762868