CLV

This repository provides the implementation details for the ACL 2023 main conference paper:

Enhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona

0.Preinstallation

  git clone CLV
  cd CLV
  pip install -r requirements.txt

1.Parts that may need to be modified

In config.py:

    # mdoel_file
    output_dir = './Model'

    # evaluation_model_file
    consis_model_dir_EN = './Consis_Model/consis_model_EN.ckpt'
    consis_model_dir_ZH = './Consis_Model/consis_model_ZH.ckpt'
    
    # Data file format
    data_path = ['/Data/Persona_Dialoue_EN','/Data/Persona_Dialoue_ZH']
    
    # Whether to use learnable metrics, 
    # and if so, train the Consis_Model first
    model_metric = False

    # language switch
    data_language = 'EN'

2.Data file format

See the samples in the Data folder.
Sample:

i have amazing children and grandchildren.i can sew my own clothes.i had cancer but its gone now.i am retired and living the great life.i do not have a smartphone.
<|endoftext|>I love iphone! i just bought new iphone!<|endoftext|>Thats good for you, i'm not very into new tech<|endoftext|>
I am a college student and i am a college student<|endoftext|>

3.Running

Setiing config.py,
Then run clv.py.
It will be tested and evaluated automatically.

4.Training Consis_Model(if you need)

Running Consis_Model.py;
Modify the model name according to consis_model_dir_EN or consis_model_dir_ZH;
Execute 3.Running.

MISC

Build upon 🤗 Transformers.