This repository contains the codes and datasets for paper "Exploiting Pseudo Future Contexts for Emotion Recognition in Conversations".
We propose to generate a pseudo future context for an utterance which serves as external conversational knowledge. Furthermore, a framework is proposed to jointly exploit multi-contexts, including historical contexts, historical speaker-specific contexts, and pseudo future contexts.
- run.py : Main function to run the codes.
- arguments.py : Containing all arguments we use.
- datasets : Containing four ERC datasets, including IEMOCAP, DailyDialog, EmoryNLP and MELD.
- models : Containing simple implementations of models.
- utils : Containing tools related to data access and processing.
- scripts : Providing some scripts for training and testing, and some checkpoints for quick start.
- roberta-base/large : To run the codes, one should download the pre-trained RoBERTa models from huggingface.