nlp-transfer

Borealis AI Internship Project (Summer 2018)

This is Yunshu's Borealis AI internship project Emperical study on transfer learning for text classification**

CNN text classification is modified from https://github.com/dennybritz/cnn-text-classification-tf and https://github.com/alexander-rakhlin/CNN-for-Sentence-Classification-in-Keras

RNN text classification is modified from https://github.com/roomylee/rnn-text-classification-tf

FastText text classification is modified from https://github.com/brightmart/text_classification

Sentence similarity measurement is modified from https://github.com/nlptown/nlp-notebooks/blob/master/Simple%20Sentence%20Similarity.ipynb

**By: Yunshu Du

Sept, 2018**

Requirements

• Python 3

• Tensorflow > 1.7

• Numpy

Training

Train once (in default config):

python3 ./train.py

Train with all config combinations using CNN model:

StackExchange Data

./run_se_cnn.sh

SICK Data

./run_sick_cnn.sh

Can also run RNN and FastText in similar way (.sh scripts)

check FLAGS defined in train.py for detailed description of configurations

Sentence similarity

Under folder "sentence_similarity"

Compute similarity using Universal Sentence Encoder

python3 USE.py

Compute similarity using other baseline methods

python3 sent_sim.py

References

• Convolutional Neural Networks for Sentence Classification

• A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification