persona

persona copied to clipboard

Open domain persona-aware dialogue generation

This codebase implemented four papers about open domain persona-aware dialogue generation, Code quality is low, just can be used for experiments:

1. Improvement of a dedicated model for open domain persona-aware dialogue generation (https://arxiv.org/abs/2008.11970):
   AttentionRoutingPlus/
   (Please contact [email protected] for the PersonalDialog dataset)

2. A Pre-training Based Personalized Dialogue Generation Model with Persona-sparse Data (http://arxiv.org/abs/1911.04700):
   AttentionRouting/
   (Please contact [email protected] for the PersonalDialog dataset)

3. Personalized Dialogue Generation with Diversified Traits (http://arxiv.org/abs/1901.09672):
   PersonalityTraitFusion/
   (Please contact [email protected] for the PersonalDialog dataset)

4. Assigning personality/identity to a chatting machine for coherent conversation generation (http://arxiv.org/abs/1706.02861):
   AssignPersonality/
   (Personality Assignment dataset: http://coai.cs.tsinghua.edu.cn/hml/dataset/#AssignPersonality)

The first two papers code tested on ubuntu16.10 and python3.6.10.

The last two papers code still have problems for training , don't use them.

Install

git clone https://github.com/ghosthamlet/persona.git

cd into the corresponding paper code directory, then:

pip install -r requirements.txt

Data prepare

for datasets of the first two papers, segment dialog sequences to characters separated by space, randomly extract 100000, 20000, 20000 sessions for three datasets without replacement: train_char.txt, valid_char.txt and test_char.txt, place them into datas directory of corresponding paper code.

Training

in the corresponding paper code directory:

python trainer.py --config_file configs/default.yaml

Evaluation

in the corresponding paper code directory:

python evaler.py --config_file configs/evaler.yaml