I have thousands of comments and I used Hugging Face's bart-large-mnli (https://huggingface.co/facebook/bart-large-mnli) to classify the comments into several topics that I entered in to the codes (i.e., candidate_labels) (zero-shot classification). It seems that the accuracy rate of correct labeling is not high sometimes. I wonder whether it is due to the model I chose. Except for the bart-large-mnli, is there any other model that can give more accurate labeling? Or does anyone have any suggestions on how to improve the accuracy rate.