search

TheophileBlard / french-sentiment-analysis-with-bert

How good is BERT ? Comparing BERT to other state-of-the-art approaches on a French sentiment analysis dataset
MIT License
146 stars 35 forks source link

Have you compared the results with FaluBERT? #25

Open LeMoussel opened 2 years ago

LeMoussel commented 2 years ago

Have you compared the results on the analysis of French sentiment with BERT, based on FlauBERT?

Rem: The CamemBERT algorithm seems to have learning problems with the Tensorflow library: Github Issue

TheophileBlard commented 2 years ago

Hi @LeMoussel !

No, I did no try to finetune FlauBERT. This model came a bit later and I was already satisfied with the results of CamemBERT. I did not experienced any issue with the TF implementation, as my validation nicely converged to >97%.

I however compared the results they reported in their paper with those of my model in this section. On their FLUE task, my CamemBERT model (trained on movie reviews) performs better than FlauBERT, even though their model was specifically trained for this task. So I would say that both models have comparable performance, it's more about the dataset they are trained on.