Inspiration

Tumblr user-generated content.

What it does

Uses neural networks and AI techniques to process a series of text in the form of forum posts in order to generate similar content for a bot-controlled Tumblr blog.

How we built it

We used a Python browser emulation framework to scrape information of internet subcultures. We wrote custom cleanup script to sanitize the training data. The training data was analyzed using a multi-layer recurrent neural network for word-level language models in Python using TensorFlow.

Challenges we ran into

Processing training data in the time allotted for the event, computationally processing data of this size takes a substantial amount of time, which became a challenge of balancing this expense with development hours.

Accomplishments that we're proud of

Successfully extracting forum content from a wide variety of website layouts. Utilization of a recurrent neural network for word level language models, application of amazon web services to run the neural network for efficient processing of training data.

Built With

Share this project:

Updates