search

Patrick-Ni / KnowEE

Code for EMNLP2023 “Multi-Source Multi-Type Knowledge Exploration and Exploitation for Dialogue Generation”
MIT License
3 stars 0 forks source link

About the Knowledge Exploration #1

Open XiangyanChen opened 4 months ago

XiangyanChen commented 4 months ago

I saw a different type of knowledge intended to load in main.py, including emotion, topic, persona, etc, but I did not find datasets to generate this knowledge. Would you like to share these dataset with me? Like "go_emotions", "plm_knowledge/preprocess_wow_train.pkl" in Persona and "atomic/v4_atomic_all.csv".

XiangyanChen commented 4 months ago

I tried the best to find these datasets exploring knowledge, but I still can't find the WOW dataset "data/plm_knowledge/preprocess_wow_train.pkl".

Patrick-Ni commented 4 months ago

Hi! Sorry for being late to reply to you. I can provide the data, but due to its large size, downloading from my server will take some time, so the update might not be ready until tomorrow or the day after. I will upload this portion of data as soon as possible and reorganize the code to keep it aligned with the paper. If you have any other questions, feel free to ask!

XiangyanChen commented 4 months ago

Thanks for the reply. Since I don't have the WOW dataset to explore knowledge, I tried to run the ablation study experiment with Dailydialog, but without WOW knowledge in the "main.py", and I got the error "assert len(utters) % 2 == 1". I think this error also will occur in the generation task(--gen). I'm wondering if these datasets used in response generation need to be preprocessed.