Financial Sentiment Analysis with FinBERT & HuggingFace + Analyzing Model Predictions w/ W&B Tables
🚀Hey everyone, and in this video we'll be looking at financial sentiment analysis with FinBERT!
To be more specific, we will perform inference on a Kaggle dataset made up of stock market news headlines using a FinBERT (Financial BERT) NLP model implemented with HuggingFace. The model will output activations for three classes: positive, negative or neutral. Those relate to how a given headline is likely to affect a given company's stock price according to the FinBERT model.
Then, after performing inference on the dataset on Google Colab, we will log the predictions to Weights & Biases and analyze them using W&B Tables, a tool for visually exploring tabular data. We'll perform general analysis of the natural language processing model predictions and look at whether certain words bias FinBERT to always output certain predictions.
---
Links
📍 FinBERT x W&B Google Colab Notebook: http://wandb.me/finbert-colab
📍 Blogpost version of the video: http://wandb.me/finbert-report
📍 My dashboard with the W&B Table from the video: https://wandb.ai/ivangoncharov/FinBER...
📍 Tables docs: https://docs.wandb.ai/guides/data-vis...
📍 Kaggle dataset: https://www.kaggle.com/miguelaenlle/m...
---
⏳ Timestamps ⏳
00:00 Intro
1:04 What is FinBERT?
1:38 Google Colab notebook
2:37 Analyzing model predictions w/ W&B Tables
3:35 Checking if a certain word biases the FinBERT model
6:17 Plotting FinBERT model predictions
10:00 Outro
---
Follow Ivan:
👉 Twitter: / ivangrov
👉 YouTube: / ivangoncharovai
Get started with W&B: http://wandb.me/intro
Follow us:
Twitter: / weights_biases
Linkedin: / weights-biases
Thanks for watching! If you have any questions, please don't hesitate to ask! If you have any suggestions, please don't hesitate either! We love hearing from the community and look forward to seeing you join our community!