/tweet_sentiment_analysis

Primary LanguageJupyter Notebook

Tweet sentiment analysis (AI Tinkerer Competition)

Applied two different approaches

  1. "classification" approach Added a MLP head to T5 model for classification.

Result: 90.7% accuracy (at step 350) Weight and Bias training report: https://api.wandb.ai/links/minki-jung/ciihjpfi


2. "text-to-text" approach Use text-to-text approach using a prefix. References this paper: https://arxiv.org/pdf/1910.10683v4

Result: 90.662% accuracy (at step 375)

Weight and Bias training report: https://api.wandb.ai/links/minki-jung/ihrsvww4


Dataset used: https://huggingface.co/datasets/zeroshot/twitter-financial-news-sentiment


note

pip install runpod requests datasets transformers numpy evaluate wandb accelerate scikit-learn nltk absl-py rouge-score