Poli-Speech

A Character-Based Text Generation Model w/ LSTM

Comment

Inspiration

We want to see the key features behind presidential speeches from both sides of the political spectrum. Along the way, we wondered what a generalized speech would be like from either party - Republican or Democrat. So, we looked to machine learning to answer this question.

What it does

This system uses an LTSM neural network to learn features of democrat and republican speeches, and generate speeches in republican or democrat style depending on which one the user requests. We also use an Amazon Alexa Echo to read the requested speech.

How we built it

  • LTSM model: Keras, Python

  • Data scraping: BeautifulSoup, Python

  • Server: Amazon EC2 (AWS), Flask

  • User Interfaces: React, HTML, CSS, Amazon Alexa Echo

Challenges we ran into

  • training and refining models for machine-generated speeches

  • integrating text-generation backend and UI

    Accomplishments that we're proud of

  • generating models for republicans and democrats, and various individual candidates

  • completing linkage between backend and frontend

    What we learned

  • scraping data effectively

  • machine learning for features of speech

  • deployment to servers

  • user interface development and linkage to backend components

What's next for Poli-Speech

  • improve neural network and quality of generated speeches

  • improve user interface design and ease of use

Share this project:

Updates