Sentiment Analysis API

This is an inference engine that exposes API endpoints which can be used to classify/analyze different sentiments from raw texts.

How to start

First of all, make sure all the dependencies are installed. To do that, make a virtual environment by executing the following command python -m venv venv Now to switch into the environment, if you're using windows then execute the command .\venv\Scripts\activate or if you're on linux or using git bash, then try source venv/bin/activate.

After doing this, run pip install -r requirements.txt. With this command all the required dependencies will be installed. Be patient, it will take some time though.

Using the app: To use the application in the local machine run python app.py. Then you can click on the link shown in the terminal to access the application with the url in the browser. You can now check the app with applications like curl or you can go with modern gui applications like postman.

You can play with the app using the /docs endpoint. There you can test the app on the fly.

Now again, to have a wholistic view of the project you can run pytest tests on the root directory of the project and again if you want to check the backend application and the API endpoint just run pytest tests/test_routes.py and to check the model inference in itself run pytest tests/test_model.py.

Technology Stack

This whole app is built with the following packages:

• SetFit
• FastAPI

SetFit is used for the AI backend. FastAPI is used to expose the inference capability for user consumption.

Dataset Description

To fine tune the AI model for our usage, it was needed to train the base model with a custom dataset. And for that purpose I used this dataset. I had to do some tweaking to get it working as we expect.

Dependencies

In terms of dependencies, we needed some external python modules which I have used extensively in this project. uvicorn is used as a server, pytest is used to have the unit tests done.

Implementation Details

First of all, we need the model to work as required. For that, I have managed a dataset that include all the needed labels such as positive, negative and neutral. As soon as I have found a suitable dataset I went to Google Colab to perform EDA and data preprocessing. With the help of huggingface datasets library I organized the data accordingly and initiated training. After training, as the moment I got the trained model, I hoped on the VS Code on my local machine and implemented the backend application. You can check the resource directory above to have a look on how the training is done.

Additional Notes( Please Read Carefully )

• Make sure you have all the necessary dependencies.
• Make sure you are on virtual environment while doing both spinning up the server and testing the endpoints.

Unable to use it?

Please file an issue above. Will try to resolve as quickly as possible.