Model Card for DeBERTa-v3-base-ECE

This is DeBERTa-v3 fine-tuned for Emotion Cause Extraction (ECE) task. For input text i.e. a sequence of tokens containing a situation with emotional coloring, it is necessary to determine the subset of which tokens justify the emotional state of the speaker. Formally speaking, it is convenient to look at the problem as a binary token classification, where one means that the corresponding token belongs to the desired subset.

Training

Code use to train this model avaliable on my GitHub

Evaluation

Has following results on EmoCause and EmpatheticDialodues:

Accuracy	Top-1 Recall	Top-3 Recall	Top-5 Recall
0.59	0.249	0.623	0.806