Inspiration

We wanted to create tool that used a model to solve an industry relevant issue. We ran across Name Entity Recognizers, and we decided to make a scaled-down version of an NER for online stores.

What it does

It model that takes in training data with product listing titles and categories, and then is able to assign (or predict) the category of an unlabeled product listing title.

How we built it

We got our data from Kaggle, and then we spent some time cleaning up the data for our purposes. Then we set experimented with a Neural Network using TensorFlow, but we ended up settling for a Naive Bayes model from SKLearn.

Challenges we ran into

We knew next to nothing about Machine Learning, so we spent a lot of time learning the libraries and technologies. We also had to switch our initial approach from a Neural Network since it didn't deliver good results.

Accomplishments that we're proud of

We built a functioning model with high accuracy. This was better than our MVP. We are proud that we managed to implement the different approach than we originally intended, and that we were able to learn new skills for this project

What we learned

We learned to use pandas, cleaning data, experimented a little with TensorFlow and SKLearn. We learned the process of making a model with ML. We also learned to use Jupyter Notebooks and Google Colab for Cloud-based collaborative programming

What's next for Listing Title Categorizer

We mean expand this into a full NER, which will be able to label individual nouns for Product Listing and label them into more specific noun categories (like "brand", "product name", "color", "gender", "size", etc)