Classify Duplicated Quora Question Pairs

The Kaggle competition

• deep_learning/: Deep neural networks which extract feature vectors from Quora questions.
  • cnn.py is a convolutional neural network.
  • lstm.py is a recurrent neural network.
• feature_engineering/: 60 hand crafted features on a pair of documents, Quora questions in this case. xgboost decision trees ingests the features and classify whether the two questions are similiar. plots.ipynb shows the feature importance of trained xgboost trees.
• matrix_fact/: Matrix Factorization based models to extract feature vectors from Quora questions, such as LSA.