We want to test the capacity of ChatGPT on Affective Computing. This repository contains the result of Dialogue Act Classification(DAC) for now and we will update more results in few weeks.
| Model | Acc | weighted-F1 |
|---|---|---|
| Co-GAT | - | 79.4 |
| ChatGPT, oneshot | 0.67 | 0.65 |
| ChatGPT, oneshot+prompt-engineering | 0.71 | 0.70 |
| ChatGPT, fewshot | 0.73 | 0.72 |
| ChatGPT, fewshot+prompt-engineering | 0.74 | 0.73 |
-
cal_prf.py
input: two label files which are chatGPT prediction and golden answer fileoutput: the classification report using sklearn -
check_utts
input: output file of chatGPT and golden answer file
output: the indexes of dialogue which are obviously wrongly annotated.
-
dd_label_extract.py: extract the labels from output file of chatGPT into a label file.
-
dd-dialogue-act.py: sample 64 dialogs in daily dialogue and transform the data format.
-
main.py: use ReverseChatGPT API to use chatGPT