huggingface / setfit

Efficient few-shot learning with Sentence Transformers
https://hf.co/docs/setfit
Apache License 2.0
2.24k stars 222 forks source link

Regarding expanding the label column #521

Open dineth99-bit opened 6 months ago

dineth99-bit commented 6 months ago

Using the sentence transformers you have provided I tried different variations for the aspect model and the polarity model and obtained good results. However I am interested in two more things which I tried and still couldn't achieve.

1) For ABSA tasks rather than three labels I wanted to expand it to 5, adding the more positive and more negative parts. We manually annotated our own dataset for this as for the format of the Laptop and restaurant 14 datasets you have provided. But it did not give the expected results. The polarities cannot be expanded to 5. How can I achieve this ? Can you guide on this matter.

2) Instead of the accuracies for the Polarity and aspect extraction separately I want an collective accuracy. A single accuracy that represents both. How can I do that?