First go to preprocess folder to generate datasets suitable for model training.
Then go to nmt-tf-1.4 folder to train the seq2seq model and then generate descriptions.
Finally go to postprocess folder to combine the descriptions from different topics and then generate the complete description for each property.