m-colombo / nlp-quora-duplicate-question Goto Github PK

The output of a project for the university: a Tensorflow implementation that faced the task posed by the Quora Question Pairs dataset, pubblicated along with the Kaggle competion. The goal is to build a classifier ables to classify whether two questions, written in english, have the same meaning. Our approach relies on pre-trained word vectors, specifically GloVe2, and deep neural network models. Two architecture have been developed and assessed side-by- side, the first in the flavour of a sentence encoder, the second more inspired by neural reasoner systems.

All the models are encoded in main.py, check on top of the file for the supported flags.

Use prepare_data.py to preprocess the data and prepare them for the training phase. The script expects to find the original files in the data/original folder, you can download them from Kaggle Check the supported flags on top of the python file.

Check the project report for more details.