Machine vs humans – predicting the Oscars with AI.

With the Oscar race heating up ahead of the awards on Sunday, there’s been plenty of debate in the office on who will take home the trophies.

As big fans of all things data, we naturally wanted to see if we could use AI to predict the results of Hollywood’s hottest awards night. So, we set a challenge to our data science team here at Found, to use Social Media and publicly available data sources to train an AI to predict this year’s Oscars. Pretty neat, right?

The machines aren’t quite at the point of achieving full sentience and destroying us all Terminator style, but they can now open doors in a chilling Black Mirror-like manner. So why not find out if an AI can predict the Oscars better than our in-house movie expert Lily and the pros at the bookies?

Our Machine #Oscarbot the AI

To train the data model, our Data Scientist James gathered data from a combination of buzz and sentiment on Twitter, ratings on IMDB, previous wins from Datahub.io and wins from other 2019 awards. The deep learning model, when tested on ‘unseen’ data (a subset of the historical Oscar data not seen by the machine during training), was able to correctly predict the winners with a 93% accuracy.

Here are the AI Oscarbot’s choices for the top categories:

  • Best picture – Roma
  • Actor in a leading role – Rami Malek
  • Actress in a leading role – Glenn Close
  • Actor in a supporting role – Mahershala Ali
  • Actress in a supporting role – Emma Stone
  • Writing (adapted screenplay) – Can You Ever Forgive Me?
  • Writing (original screenplay) – Roma
  • Best director – Alfonso Cuaron
  • Animated feature film – Isle of Dogs

So a few unusual choices there from #Oscarbot – notably Emma Stone for ‘Actress in a supporting role’ who the bookies (and our film buff) see as a rank outsider at 18/1 – is this an indication that it’s gone a bit Johnny 5 and going to get short circuited by the competition, or is it a sign that it’s seen something in the data that the others haven’t?

Scroll down for Oscarbot’s full list of predictions.

Our Expert – the agency film buff Lily

Data Analyst Lily is a self-proclaimed film nerd who jumped at the chance to beat the machine with her predictions. Here’s Lily’s take on predicting the results:

“While the Golden Globes are voted on by critics, the Oscars are voted on by a group of industry peers. You have to have won an Oscar or been sponsored by two members to join and must have been active in the industry in the last 10 years or be a lifetime member (been active for 30 years or won an Oscar) in order to vote. 72% of the Academy are male, 87% are white, and actors make up the largest proportion of the voting block. This means that winning films have a very specific demographic to appeal to and actor-centric movies have a huge advantage.

However, due to controversies in recent years (#OscarssoWhite especially), the academy is working to change the perception of them as out of touch. This is pertinent as the rating and viewership has been dropping over the last few years.

Anything too subversive (Get Out), strange or weird (Phantom Thread) or a so called ‘genre’ film (Mad Max) is still unlikely to take home the main awards. Traditionally, genre films have dominated technical awards and respectable dramas with a white man as the lead have taken the Acting and Best Picture awards. This has begun to change in recent years with ‘Moonlight’ and ‘The Shape of Water’ winning in the last two years.

The most difficult category is Actress in a Supporting Role. Emma Stone won the SAG award, Rachel Weiss the BAFTA, and Regina King won the Golden Globe and the Critics’ Choice award. I think this is likely going to Regina King, even though If Beale Street Could Talk was snubbed a Best Picture nod. But don’t be surprised if I am wrong here.”

  • Best picture – Roma
  • Actor in a leading role – Rami Malek
  • Actress in a leading role – Glenn Close
  • Actor in a supporting role – Mahershala Ali
  • Actress in a supporting role – Regina King
  • Writing (adapted screenplay) – BlacKkKlansman
  • Writing (original screenplay) – The Favourite
  • Best director – Alfonso Cuaron
  • Animated feature film – Spiderman: Into the Spiderverse

The Bookies

In order to add a further element to our experiment, we’ll compare our results to the bookies’ predictions. The following were correct as bookies’ favourites as of 20th February 2019 according to oddschecker.com:

  • Best picture – Roma
  • Actor in a leading role – Rami Malek
  • Actress in a leading role – Glenn Close
  • Actor in a supporting role – Mahershala Ali
  • Actress in a supporting role – Regina King
  • Writing (adapted screenplay) – BlacKkKlansman
  • Writing (original screenplay) – The Favourite
  • Best director – Alfonso Cuaron
  • Animated feature film – Spiderman: Into the Spiderverse

As you can see above, the bookies are very much in line with our expert Lily’s predictions. However, when it comes to the longer list of categories, there are some notable differences in the more technical awards. This means we should have a clear winner between the three come Monday (unless The Favourite sweeps the board, and everyone is proven wrong).

Also notable is that our #Oscarbot has chosen different winners for the supporting actress, animated film, cinematography and writing categories. Will there be a few unexpected upsets on Sunday night? Our AI seems to think so.

Who will win? Can AI beat our movie expert? Will Lady Gaga cry? Stay tuned for our follow-up after the big night.

(Disclaimer – do not hold us responsible if you lose cash at the bookies!)

The post Machine vs humans – predicting the Oscars with AI. appeared first on FOUND.