HackBU2020

A Candidate Like You

Adiel Felsen, Brendan Klayman, Nick Pellegrino, Richard Quinlivan

Description

A Candidate Like You is a multi neural network website aimed at finding you the presidential candidate that is the most like you, however you determine that.

Finding most like-minded candidate:

  • A ChatBot prompts conversation with the user about their political views
  • Two neural networks analyze speech patters, language use, and vocabulary to create an opinion profile on the user
  • Once enough data has been gathered, the user is shown the politician that best matches their custom opinion profile

Finding most look-alike candidate:

  • A Convolutional Neural Network (CNN) is trained on 135 images of each presidential candidate with the goal of classifying the candidate
  • The CNN is tested on 15 images of each candidate
    • Achieved ~85% training accuracy and ~62% testing accuracy
  • The CNN outputs the candidate who looks like you the most, and a saliency map that represents what the neural network is looking at

Built With

Share this project:
×

Updates