Sentiment Analysis

Sentiment analysis neural network trained by fine-tuning BERT, ALBERT, or DistilBERT on the Stanford Sentiment Treebank.

Setup Sentiment Analysis

Clone Sentiment Analysis, update lists of packages, install Pyton 3.9

git clone https://github.com/barissayil/SentimentAnalysis.git
cd SentimentAnalysis
sudo apt-get update
sudo apt-get install python3.9

Verify that you have Python 3.9.10

python3.9 --version

Create virtual environment with Python 3.9.10 binary, activate it

python3.9 -m venv env
source env/bin/activate

Install necessary packages

pip install -r requirements.txt

Test it and verify that it passes all tests

python -m pytest

Use Sentiment Analysis with my model

Evaluate

python evaluate.py

Analyze your inputs

python analyze.py

Run server, and verify that you get back 99% positive

python server.py

curl localhost:5000 -G -d text=good

Train your own model and use Sentiment Analysis with it

Train (i.e.fine-tune) BERT

python train.py --model_name_or_path bert-base-uncased --output_dir XXX --num_eps 2

bert-base-uncased, albert-base-v2, distilbert-base-uncased, and other similar models are supported.

Evaluate

python evaluate.py --model_name_or_path XXX

Analyze your inputs

python analyze.py --model_name_or_path XXX

Run server

python server.py --model_name_or_path XXX