abstractive summary and title generation

Corner stone seq2seq with attention (using bidirectional ltsm )

the model

• uses an encoder with multi layer RNN with LSTM
• Decoder is built using bahadau attention model
• Inference was cstom built by hand

Data is

• arxivscraper research papers

Word2Vec

• Conceptnet Numberbatch's (CN) embeddings, similar to GloVe, but probably better (https://github.com/commonsense/conceptnet-numberbatch)