In this project we compare three summarization models on the WikiHow dataset.
Each of the model is uses a different embedding to encode the sentences. The results are compared and presented in the report.
Embeddings used are
- Word2Vec
- GloVe
- BERT
See word2vect_README.pdf for word2vec model
See GloVeReadMe.txt for GloVe model
The bert_encoder_embedding.py file contains the code to train and test the model. The outputs generated are stored in without_stopwords_output.txt. attention.py file was obtained from the (https://github.com/thushv89/attention_keras/blob/master/src/layers/attention.py) as keras does not have an inbuilt attention layer.
Original summary: how to stop racing thoughts in the middle of night
Predicted summary: how to stop anxiety at night
Original summary: how to make giant cookie
Predicted summary: how to make chocolate chipcookie
Original summary: how to avoid an pylori bacte-rial infection
Predicted summary: how to prevent bacterial infection
