DialogRPT-depth

DialogRPT-depth / README.md
system
Update README.md 4431e67
1
# Demo
2
3
Please try this [➤➤➤ Colab Notebook Demo (click me!)](https://colab.research.google.com/drive/1cAtfkbhqsRsT59y3imjR1APw3MHDMkuV?usp=sharing)
4
5
| Context | Response | `depth` score |
6
| :------ | :------- | :------------: |
7
| I love NLP! | Can anyone recommend a nice review paper? | 0.724 |
8
| I love NLP! | Me too! | 0.032 |
9
10
The `depth` score predicts how likely the response is getting a long follow-up discussion thread.
11
12
# DialogRPT-depth
13
14
### Dialog Ranking Pretrained Transformers
15
16
> How likely a dialog response is upvoted 👍 and/or gets replied 💬? 
17
18
This is what [**DialogRPT**](https://github.com/golsun/DialogRPT) is learned to predict.
19
It is a set of dialog response ranking models proposed by [Microsoft Research NLP Group](https://www.microsoft.com/en-us/research/group/natural-language-processing/) trained on 100 + millions of human feedback data. 
20
It can be used to improve existing dialog generation model (e.g., [DialoGPT](https://huggingface.co/microsoft/DialoGPT-medium)) by re-ranking the generated response candidates.
21
22
Quick Links:
23
* [EMNLP'20 Paper](https://arxiv.org/abs/2009.06978/)
24
* [Dataset, training, and evaluation](https://github.com/golsun/DialogRPT)
25
* [Colab Notebook Demo](https://colab.research.google.com/drive/1cAtfkbhqsRsT59y3imjR1APw3MHDMkuV?usp=sharing)
26
27
We considered the following tasks and provided corresponding pretrained models.
28
29
|Task | Description  | Pretrained model |
30
| :------------- | :----------- | :-----------: |
31
|  **Human feedback**  |  **given a context and its two human responses, predict...**|
32
| `updown` |  ... which gets more upvotes? | [model card](https://huggingface.co/microsoft/DialogRPT-updown) |
33
| `width`| ... which gets more direct replies?  | [model card](https://huggingface.co/microsoft/DialogRPT-width) |
34
| `depth`|  ... which gets longer follow-up thread?  | this model |
35
|  **Human-like** (human vs fake) | **given a context and one human response, distinguish it with...** |
36
| `human_vs_rand`| ... a random human response  | [model card](https://huggingface.co/microsoft/DialogRPT-human-vs-rand) |
37
| `human_vs_machine`| ... a machine generated response  | [model card](https://huggingface.co/microsoft/DialogRPT-human-vs-machine) |
38
39
40
41
42
### Contact:
43
Please create an issue on [our repo](https://github.com/golsun/DialogRPT)
44
45
### Citation:
46
```
47
@inproceedings{gao2020dialogrpt,
48
    title={Dialogue Response RankingTraining with Large-Scale Human Feedback Data},
49
    author={Xiang Gao and Yizhe Zhang and Michel Galley and Chris Brockett and Bill Dolan},
50
    year={2020},
51
    booktitle={EMNLP}
52
}
53
```
54