The Spoiler started as the final project for CSCI 5622, Machine Learning, at the University of Colorado Boulder. The version you see here is a Random Forest Classifier trained on approximately 150 movie trailers, with the results (happy ending, death of a protagnist and romantic success) annotated by hand (therefore there is room for error and subjectivity).

The classifier outputs a binary classification (yes or no) with a probability of that answer. Results are displayed with a confidence score, which is defined as that probability minus 50%, times two; ie, a 60% probability of "yes" means there was 40% probabilty of "no", so we display a confidence of 20%.

In the future we plan to add the ability to crowdsource additional movie results to improve accuracy. In addition, we would like to add character specific results (ie, does Character X die?), as well as use additional features in the trailer.

For questions/comments/suggestions, contact us at spoiler@loudsound.org.