kartiksethi/sarcasmdetection

A deep learning implementation to detect sarcastic texts in short textual corpora.

SarcasmDetection

A Bi-LSTM based attention model implementation to classify texts as sarcastic or not sarcastic.

Dataset used: sarcasmV2 dataset

Steps to be followed to make the code work:

• Install latest version of Keras and TensorFlow
• Install the other python libraries: numpy, pandas, scikit-learn, matplotlib, etc.
• data_preprocessing.py contains the script to pre-process and load the data.
• All the different models are stored in models.py.
• sarcasm_classifier.py is the main file and it contains the code for training and testing the models.
• Command line arguments for sarcasm_classifier.py:

  [--epochs]: Number of sweeps over the dataset to train
  [--lr]: Learning Rate for the optimizer
  [--optimizer]: Type of optimizer
  [--batchsize]: Number of sequences to process at a time
  [--model-type]: Type of model to run from models.py
  [--kfolds]: Defines whether we want to do k-fold cross-validation or not
  [--save]: File path to save the final model
  [--model]: Give a file path of model to test
  [--transfer-learning]: File path to pre-trained glove embeddings
  [--trained-embeddings]: File path to word embeddings of words from sarcasmV2 dataset
  [--data-path]: File path to dataset file
  [--model-summary]: To enable or disable model summary

Note: Please download the GloVe embeddings (glove.6B.zip) and store them in appropriate directory before proceeding with transfer learning. One can use their own pre-trained word-embeddings as well.