Are you guys using the nlp.pipe() for pre-processing the text as suggested in the starter notebook? I am using that and tried different models, but can never surpass the 0.7 acuracy barrier
Try out the BERT based model from the notebook I just shared: https://www.aicrowd.com/showcase/bert-for-emotion-detection
Should get you above 0.8 I think. Let me know in case of any issues.