Review Sense

Inspiration

We noticed that there are many Amazon reviews with users rating anywhere between 1-5 stars but not every review is acknowledged or used to improve the products for a better user experience. Our goal was to create a platform on which companies can view the reviews they receive and be given suggestions on actions they can take to improve the user experience.

What it does

Our platform analyzes reviews of electronic Amazon products to provide companies/sellers a clear image how well or bad their products are being received by consumers. Instead of manually sifting through thousands of reviews for a single item, companies can now get summarized data on a given product in regards to the most common complaints or compliments of a product. This makes it easier for sellers to identify what aspects of a product need improvement and what is already doing well. When an item receives are great amount of negative reviews, it is flagged so that companies know to give priority to improving those products.

How we built it

We built the platform using Python and Supabase for the back-end, with Python being used for analyzing and training the data model while Supabase was used to store the reviews as well as update new reviews that came in. For front-end, we used React and Flask to make a display to view the results in a appealing format.

Challenges we ran into

Initially, we had issues with cleaning our data set, largely due to the size of it. In addition, we switched datasets and ideas a couple of times as we understood the prompt which took away some time from our work time. We also had a hard time finding pre-trained models to label our data qualities.

What we learned

For the back-end, we learned about model tuning, featurization, new types of models and how to extract data from them. For the front-end, we learned how to use React and Flask as well as how to connect it to the back-end.

What's next for

The next steps we would take with this project is to implement it for a larger dataset. Currently, we have it implemented mainly for headphones. It would be nice to cover more electronics as well as the other categories of reviews that Amazon receives within the pages of the interface. Another step would be to implement it to live database using Amazon's API.