Flaming Strainer

A machine learning based real-time comment scoring and sentiment analysis service to host better conversations.

Comment

Inspiration

  • Hostile online interaction that involves insulting messages, or flames, between users, needs to be dealt with using a continuously learning solution (ML model working on a stream data)

What it does

  • Real-time ‘Flame Strainer’!
  • Host better conversations!
  • Scoring the impact of a comment in a conversation
  • Flexible parameters setting - platform administrator

How I built it

  • Multi-class classification using LSTM + CNN implemented with keras with tensorflow backend
  • Sentiment Analysis using Google Cloud Language API offered via Google Cloud Platform
  • Training and deployment of the trained model for prediction tests on the Google Compute Engine

Challenges I ran into

  • Issues transferring huge word embedding files to and from Google Cloud Platform
  • Low-performance hardware; I was not able to train the deep learning model for more than 5 epochs

Accomplishments that I'm proud of

  • Ability to implement the core objective of the idea in time
  • Opportunity out other teams with Android and Cloud Deployment related issues

What I learned

  • Development using Flask and Jinja
  • Writing Consumer Client for the Google Cloud Language API

What's next for Flaming Strainer

  • Deployment of “Flaming Control” tools on real-time production systems
  • Training of the model using the live streaming data
  • Additional word and character embeddings
  • DEEPER Learning!

Built With

Share this project:

Updates