Home

Awesome

ege-RoBERTa

Learning event graph knowledge for abductive reasoning.

To this end, we involve a two stage learning process to introduce the event graph knowledge, and a variational autoencoder based model ege-RoBERTa to capture the event graph knowledge.

Pre-training Stage:

Learning Event Graph Knowledge from a Pseudo Instance Set

Finetuning Stage:

Adapt Event Graph Knowledge to the Abductive Reasoning Task

Please refer to train_anli.sh and Train_anli.py.

Model Architecture

Files for constructing model architecture is contained in the file folder onmt.