RNN semantic classification

Inspiration

I began with your standard run of the mill neural network, and only learn of recurrent neural nets after a bit of digging. After learning that they excelled in dealing with sequential data, it seemed natural to consider language processing and classification from there

What it does

Classifies text samples into either positive or negative semantic categories

How we built it

Built purely in python, and trained on first a dataset of my own making, and then a dataset of IMDb reviews from kaggle in its latest iteration

Challenges we ran into

I ran into a number of challenges with this project. Among the first of these was the size of the dataset I was working with greatly impacting speed and efficiency.

Moreover, convergence was sadly not achieved, and while I still require more exploratory work, I believe an emphasis on data cleaning/processing pipelines could certainly help.

Accomplishments that we're proud of

I’m proud of testing and workshopping several possible solutions to the issues that arose throughout the project.

What we learned

In pursuing this project I learned about key machine learning concepts like back propagation, Xavier initialization, activation functions, vector encoding, as well as a number of other concepts.