preke / DesPrompt

source code

DesPrompt

This reposity contains the source code and data for the paper DesPrompt: Personality-descriptive prompt tuning for few-shot personality recognition

Source Code:

Following the workflow above, we generate the prior verbalizers (words and their corresponding weights) by src/label_words/Prior Verbalizer Generation.ipynb;

Then, we Pre-finetuning the T5 models by src/Cohesive Pre-finetuning.ipynb;

Then, in Coherent Prompt Generation, we generate the templates through src/Coherent_Prompt_Generation.ipynb; (This step is quite time-consuming, so, we also provided the templates we generated in src/templates/)

Then, we generate the Posterior Verbaliser through src/Posterior Verbalizer Generation.ipynb;

Finally, we conduct the Prompt-based fine-tuning through src/main.py

Single Sample API:

We also provide a code to inference the personality of an input single sentence with our method: /src/single_sample_api.py.

You can modify the input sentence at line 92.

Required packages:

pytorch==1.13.0

transformers==4.23.1

openprompt==1.0.1

jupyer notebook==6.4.12

Citation

If the code helps you, please kindly cite the following paper:

@article{wen2023desprompt,
  title={DesPrompt: Personality-descriptive prompt tuning for few-shot personality recognition},
  author={Wen, Zhiyuan and Cao, Jiannong and Yang, Yu and Wang, Haoli and Yang, Ruosong and Liu, Shuaiqi},
  journal={Information Processing \& Management},
  volume={60},
  number={5},
  pages={103422},
  year={2023},
  publisher={Elsevier}
}