Dialogue Meaning Representation (DMR)

This is the offical repository of the paper Dialogue Meaning Representation for Task-Oriented Dialogue Systems . It contains the DMR-FastFood dataset and code for DMR parsing and coreference resolution.

@article{hu2022dialogue, title={Dialogue Meaning Representation for Task-Oriented Dialogue Systems}, author={Xiangkun Hu, Junqi Dai, Hang Yan, Yi Zhang, Qipeng Guo, Xipeng Qiu, Zheng Zhang}, journal={arXiv preprint arXiv:2204.10989}, year={2022} }

DMR-FastFood dataset

The splits of the dataset are contained in fold dataset/dmr-fastfood. The following table shows the detailed statistics:

| | Train | Dev | Test | |--------------------|---------|--------|--------| | Dialogues | 5,585 | 710 | 899 | | Utterance | 102,843 | 13,111 | 16,889 | | Utterance/Dialogue | 18.41 | 18.47 | 18.79 | | Customer Utterance | 54,465 | 6,911 | 8,952 | | Utterance Length | 10.24 | 10.28 | 10.25 | | Utterance for NLU | 23,633 | 4,256 | 5,581 | | Reference | 6,007 | 802 | 1,039 | | Negation | 430 | 62 | 65 | | Conjunction | 11,770 | 1,499 | 1,989 | | NLU DMR Depth | 2.43 | 2.66 | 2.64 | | NLU DMR Nodes | 3.18 | 3.46 | 3.43 |

Run the code

The code of the models is contained in fold src. Please refer to the paper for the details of the models.

Install requirements

The main requirements are: - Python 3.7+ - torch 1.9.0 - transformers 4.9.1 - dgl 0.7.2

They can also be installed by: bash cd src pip install -r requirements.txt To install dgl, please follow the instructions in DGL website.

DMR Parsing model

To train the DMR parsing model, run: bash python run_parsing.py --add_role --constrain_decoding --train

After training, run the follow command for evaluation: bash python run_parsing.py --add_role --constrain_decoding --test

The code for calculating Smatch is adapted from snowblink14/smatch.

Coreference Resolution model

Run the following command to train the model: bash python run_coref.py --add_refer_edge --add_global_node --train

To evaluate, run: bash python run_coref.py --add_refer_edge --add_global_node --test

Security

See CONTRIBUTING for more information.

License Summary

The documentation is made available under the Creative Commons Attribution-ShareAlike 4.0 International License. See the LICENSE file.

The sample code within this documentation is made available under the MIT-0 license. See the LICENSE-SAMPLECODE file.