Spam Slayer

Home screen
Charts and wordcloud for ratings
Organic Reviews

Inspiration

My friend and I are a fan of this show called Silicon Valley, its a comedy show about Startup culture and Silicon Valley. In one episode, this character tricks children into writing thousands of spam reviews on Amazon from pre-written scripts. We thought this was funny but some research on google shows that this a serious problem, hence we decided to slay the fake reviews armed with machine learning.

What it does

When you are browsing any product on Amazon and want to get its organic reviews and correct rating, the user just needs to take the URL of this product and enter it into Spam Slayer website.

Once the user enters URL of any product on amazon, our application scrapes all the reviews of this product and then classifies the reviews as fake or not_fake. All the organic reviews classified as not_fake are returned to the frontend

The frontend presents the organic reviews to the user in an actionable format

How we built it

We used bs4 to scrape the product reviews from amazon
We used Google AutoML for training a text classifier for classifying fake reviews
The backend was built using FastAPI and deployed on GCP VM
The frontend was built in react

Challenges we ran into

Scraping and classifying text took a lot of time, we could not have kept the frontend waiting lest we get a timeout error. This caused us trouble.

Accomplishments that we're proud of

It was at this point we learned about background tasks and implemented them in FastAPI. This lets us run intensive tasks in the background without having to keep the frontend waiting

What we learned

We learned how to use background tasks and run compute-intensive programs in the backend. We also learnt how to connect the front end and backend together in a full-fledged application.

Using Google's APIs were new for us. We were quite surprised by how easy it is to use Google's Auto ML