Think before you type!

Team Members

Launch Site

Privacy Preserving Real Time Chat Sentiment Analysis

Sentiment analysis in group chats using Tenserflow JS.

V800 team 569ce8e07de8872a68d529e60cc6f972 1526308931

Description

With Facebook recently revealing the amount of hate/bully/abuse posts on their platform (http://www.bbc.com/news/av/technology-44128537/facebook-reveals-scale-of-abuse), controlling such posts has become a top priority. This is an effort to solve it with Real Time sentiment analysis.

Sometimes, during textual conversations/e-mail, the tone of the conversation/mail does not convey the actual feeling. I have used a pre-trained CNN network for helping users get details about tone of their chat message in real time (before actually sending the message to the receiver). This is especially helpful for non-english speakers as many times they are not able to convey their actual feeling. All the Machine Learning/Prediction happens in the browser itself. I have not used any 3rd party sentiment analyser API. So the user's message/data WILL NOT leave his/her system which is the most critical point.

NOTE1: The CNN model is trained on IMDB reviews and classifies text as positive/negative. It works much better with long sentences.

NOTE2: Due to an issue with latest firebase SDK, there are intermittent login issues. If you are able to login but cannot post chats/messages, in that case, please open in INCOGNITO mode or clear cache and proceed.

Instructions

Just sign up with any e-mail/password/display name and start chatting with others.

View the screencast:
https://www.youtube.com/watch?v=MimH33IXeYE

Built With

Angular, Tensorflow JS, Firebase and AngularFire 2 (shoutout to Wes Doyle https://www.youtube.com/channel/UCfniixfhHqpIGbU7z2JCNJw whose videos really helped me)

Feedback / Ratings (11)

All Feedback