|
--- |
|
datasets: |
|
- tweet_eval |
|
language: |
|
- en |
|
pipeline_tag: text-classification |
|
--- |
|
|
|
# Sentiment Analysis with BERT |
|
|
|
This BERT-based model for sentiment analysis was created by me as a student completing Vanderbilt Data Science Institute's AI Summer Course in 2023. It serves as an introductory example of fine-tuning a pretrained model for a downstream task. |
|
|
|
## What I Learned |
|
|
|
- Leveraging transfer learning instead of training a model from scratch |
|
- Fine-tuning a pretrained model on a downstream dataset |
|
- Implementing optimizations like learning rate scheduling |
|
- Evaluating models using relevant metrics like accuracy |
|
|
|
## About the Project |
|
|
|
During the program, I explored various techniques for adapting powerful large-scale models like BERT to specialized applications. As a hands-on exercise, I fine-tuned BERT using the tweet_eval dataset to classify text snippets as either positive or negative in sentiment. |
|
|
|
This model is the result of that exercise, providing my basic implementation of sentiment classification using BERT fine-tuning. While not as performant as state-of-the-art sentiment models, it demonstrates the workflow and techniques I learned around tailoring BERT and similar models. |
|
|
|
The training code is provided to allow replication and customization for other datasets. I hope this model provides a useful case study for anyone beginning their journey into fine-tuning and transfer learning with transformer models like I was! |
