Inspiration
We all love a good cult classic to put on, watch, and judge. But how well do we fare at making one? This is what the Cult Classic Calculator aims to find out.
What it does
Our goal is to determine whether a movie is a cult classic or not. A user will enter details about a fake film, and the program will return the probability of that movie being a cult classic.
How we built it
First, we scraped Wikipedia for a comprehensive list of cult classic films. We used a dataset from The Movies Database. After data cleaning, we had 10,000 movies, which is about 2 years of continuous movie watching. We vectorized all text data using BERT and trained a random forest model on model attributes title, tagline, description, runtime, and genre. Finally, we created an API using FastAPI to pass user input to the model. This model then returns a score that determines the likelihood of your built movie becoming a cult classic
Challenges we ran into
Our initial BERT model had very low precision. We resolved this by changing the parameters we were training on to mostly numerical features. Additionally, after encoding our data, we tried a linear regression model to find the likelihood of becoming a cult classic. This gave us a very low accuracy score, which we circumvented by switching to a random forest. Our API did not end up being able to post the results to the website as well, an issue that we could not fix in time.
Log in or sign up for Devpost to join the conversation.