Recently we experienced several events where a lot of us were surprised with the outcomes (eg. Brexit, US presidential elections). Most/all of our friends generally share our opinions on social media. There must be a way we can statistically assess this to help us form unbiased opinions.
What it does
Social Bubbles allows a user to input a phrase, and uses their network to visualise the social bubbles in their network. The social bubbles will also make inferences about what likely opinions your friends might have, and how likely you are to see people with different views on your Facebook Feed.
How we built it
Social Bubbles takes the facebook friendships, friends posts and their connections of a user using machine learning (K-Mediods) to cluster the posts into groups. We filter the posts by the provided phrase. We then consider the Sentiment of each post using Microsoft Cognitive Services. We finally classify the users (based on their posts assignments) to different clusters using a Bayesian Classifier.
The result is displayed finally on a D3 Web interface with useful graph statistics such as network percentage, groups interconnectedness and top works per cluster.
Challenges we ran into
Retrieving the Facebook Data Tweeking our clustering algorithm Data mining/cleaning