from Hong Kong University of Science and Technology(HKUST) Human Language Technology Center

Paper: End-to-End Recurrent Entity Network for Entity-Value Independent Goal-Oriented Dialog Learning

• Clone the repo and the dataset
• Run python REN.py --train --task=1 to begin train on task 1
• Run python REN.py --train --task=1 --record to begin train on task 1 with recorded delexicalization (RDL) data
• Use --augment to increase the dataset by partial dialog
• Use --generateRDL to generate RDL data (which was generated here)

• tensorflow 1.2
• python 2.7

• Test set 1 uses the same KB as for the train dialogs, and the same set of slots in the queries
• Test set 2 uses the different KB (with disjoint sets of restaurants, locations, cuisines, etc.), termed Out-Of-Vocabulary (OOV), but the same set of slots in the queries
• Test set 3 uses the same KB as for the train dialogs, but one additional slot for the queries
• Test set 4 uses the different KB (OOV) and an additional required slot