DB_AI

how to run the project

first create a conda environment

conda create --name sentiment-env python=3.13.2

activate the conda environment

conda activate sentiment-env

install required libraries

pip install -r requirements.txt

install onnx modules to be able to run the model on the server

conda install -c conda-forge onnx onnxruntime

then run flask server

python app.py

then go to the url

http://127.0.0.1:5000

the data analysis and model training code is in the sentiment-analysis.ipynb file

Observations & model accuracy

Vector Embedding Results

Neural Network : 71%

ML Models : 65-67 %

TF-IDF results

Neural Network : 56%

ML models : 56-65 %

final verdict

• Vector Embedding-based sentiment analysis works better than the TF-IDF method, yielding better accuracy.
• Because the dataset is limited, there is a possibility that if the dataset size is increased, then the model would be better able to grasp the sentiments more accurately.
• Overall, LLMs are much more suited for multi-class sentiment classification, as they can understand nuances present in the mail, which embedding or traditional NLP methods may ignore right away.