Sentiment Analysis API

To perform sentiment analysis on 2016 Presidential Election tweets. In my previous approach I have used Naive Bayes Classifier to solve the problem, however LSTMs(Long Short Term Memory network) provide more accuracy than traditional ML approaches( given sufficient data :) ) because of their ability to "remember" i.e to keep track of context in text sequences. Also, an increase in accuracy from 78%(using Naive-Bayes) to 87% was achieved using LSTMs.Here, I have deployed my model as REST Api so that it can easily be used in different platforms.

Training Dataset Accuracy: 87.62%

Testing Dataset Accuracy: 80.17%

Why use LSTM?

LSTMs(Long Short Term Memory networks) overcame the vanishing gradients problem in RNN and this resulted in amazing results in speech to text conversion and the advancement of Siri, Cortana, Google voice assistant, Alexa etc. They also improved machine translation, which resulted in the ability to translate documents into different languages, translate images into text, text into images, and captioning video etc.

Dataset downloaded from : link

Folder Structure:

• api/ : contains all the files needed for the flask container
• nginx/ : contains all the files needed for the nginx container
• docker-compose.yml : config file for docker-compose

How to Run :

Make sure that docker is installed on the system before starting.

1. Clone the directory

2. Build the container images (from the directory of the repo)

$ docker-compose up --build

Note : --build is only needed the first time you are running this command.

Now the container will be up and running on http://localhost .

3. Send Request

Open a new terminal and send a POST request

$ curl -H "Content-type: application/json" -d '{"_query_":"Enter Your Input Here"}' 'http://0.0.0.0:8000/predict'

How to run the API component seperately

1. Move into the api folder

$ cd api

2. Run using gunicorn

$ gunicorn -w 1 -b :8000 app:app

3. Test API endpoint from terminal

$ curl -H "Content-type: application/json" -d '{"_query_":"Enter Text Here"}' 'http://localhost:8000/predict'

Working :

Nginx works as reverse proxy and faces the outside world. It serves media files(images,CSS etc) directly from the file system.However, it can't directly talk to the python web app. It needs something that wil serve the web app with requests and gets back responses. This requirement is satisfied by Gunicorn. Gunicorn(WSGI server implementation) serves the web app with requests and gives back the responses to nginx.

The outside world <-> Nginx <-> Gunicorn <-> Web App

How to Improve Accuracy :

One approach is to use an Attention based model like Transformers. RNNs and LSTMs are more difficult to train as compared to Transformers as they require a lot more memory-bandwidth for computation. Hence they cannot utilize hardware acceleration.