This is the code repository of The Evaluation of Chinese Human-Computer Dialogue Technology (SMP2019-ECDT) Task2: Personalized dialogue. The official evaluation scripts and submission guide are released on the codalab. We achieved 3rd rank on the evaluation leaderboard. Our code is based on Seq2Seq and pretty simple, but with high scalability.
- PyTorch >= 1.0
- NLTK
- tqdm
- tensorboardX
Due to data license, all train/valid/test data should be accessed via the email smp2019ecdt@163.com
. For data preprocessing, run:
sh process_data.sh
For model training, run:
sh run_train.sh
For model testing, run:
sh run_test.sh
For evalution, run:
sh eval_bleu.sh
sh eval_distinct.sh
sh eval_ppl.sh