Text (Sentence) Classification

Text Pool (Corpus)

Transcript version of:

  • President Obama 2009 Inaugural Speech
  • 2012 Presidential Candidate Gov Mitt Romney campaign Speech

Snapshot of the Transcript versions

Game Plan

  • Every sentence extracted and labelled by the speaker's name and saved as a data-frames.

  • Both data matrix concatenated to create the corpus in data-frame format (Columns: Sentence and Speaker)

  • Text cleaning: removePunctuation(), removeNumbers(), removeSparseTerms() etc

  • Reshuffle the order of observation randomly and spilt into Training set and Test set data and pass it through different algorithms.

  • Do prediction for the test data

Classification Algorithm

  • Suport Vector Machine (SVM)
  • Random Forest (RF)
  • Decision Tree (Tree)

  • Precision: Among the total sentences predicted to be Pres.Obma’s, 85% of them are factually correct. On the other hand, 73 % of Gov Romney’s predicted sentences are factually correct.

  • Recall: 95 % of Gov Romney’s sentence are predicted correctly. Whereas, 45% of Pres. Obama’s speech sentences are managed to be predicted correctly which is below a random guess.

  • Accuracy: the selected model Random Forest had a 75 percent accuracy. This means that any randomly selected sentence has a 75 % chance of being labelled to the correct speaker.