/meed

Multi-Turn Emotionally Engaging Dialog Model (MEED)

Primary LanguagePythonCreative Commons Attribution 4.0 InternationalCC-BY-4.0

Multi-Turn Emotionally Engaging Dialog Model (MEED)

DOI

Code for paper

Yubo Xie, Ekaterina Svikhnushina, and Pearl Pu. A Multi-Turn Emotionally Engaging Dialog Model. IUI 2020 Workshop on User-Aware Conversational Agents. PDF Link.

Environment

The project was developed using the following packages:

  • TensorFlow 1.12.0
  • NumPy 1.15.4
  • spaCy 2.0.18

Files

  • model.py: implementation of the model;
  • attention.py: implementation of the hierarchical attention mechanism;
  • attn_wrapper.py: wrap the attention mechanism into the RNN cells;
  • train.py: train the model on the training set;
  • validate.py: evaluate the model on the validation set (choosing hyperparameters);
  • predict.py: evaluate the model on the test set (predict the responses).

Training Data

The training data (tokenized) can be found here.

License

See the LICENSE file in the root repo folder for more details.