An implementation of the Rush, Chopra, Weston work http://arxiv.org/abs/1509.00685
Our implementation differs in that we fix the context and summary token embedding matrix. Two scenarios are possible:
- Both embedding matrices are initialised from GloVe
- Context embedding matrix are initialised from Skip-Thought Vectors http://arxiv.org/abs/1506.06726, summary from GloVe
Helptext is available for running our code by doing:
python3 ass_train.py -h
The training corpuses are under data/
. Each JSON file contains three fields title
, full_text
, summary
. They're downloaded with scripts in download_data/
.
GloVe data needs to be downloaded and unzipped under glove/
. The code uses the first 10k most frequent tokens by default. To generate the embeddings for them,
cd glove
head -n 10000 glove.6B.300d.txt >glove.10k.300d.txt
SkipThoughts data needs to be downloaded to access the SkipThoughts embeddings (to be done).