Understanding User Comments via Sentiment Analysis

Includes analysis of a large corpus of positive and negative user comments, data cleaning, model selection, and deployment to a Flask REST API

• Nathaniel Haddad [email protected]
• Northeastern University
• Disclosure: this is a academic project

Technologies: TensorFlow, Sklearn, NLTK, Pandas, Flask, Python

---

Install

packages:

• pip3 install pickle-mixin
• pip3 install tensorflow
• pip3 install -U scikit-learn
• pip3 install Flask
• pip3 install nltk
• pip3 install pandas

run:

• train a model: (from the root folder) python comment_clf_model.py
• run the server: (from the root folder) python comment_clf_app.py
• go to the server home: http://127.0.0.1:5000/v1/api

Overview

This project represents a series of machine learning models used to identify attacks on users on Wikipedia using natural language processing. Using Scikit-learn and other packages, I built several classifiers that were able to predict whether a comment was an attack or not with a high rate of accuracy.

Notes

• Some of the resources I used:
  • https://www.kdnuggets.com/2019/01/build-api-machine-learning-model-using-flask.html
  • https://arxiv.org/abs/1912.02292