top of page
Search

Netflix's Aesthetic Visual Analysis

Updated: May 5, 2022



Making a movie could be a daunting task. It requires one to employ the right decision in different sectors like what genre to focus on, which actress to cast, which producer to hire, and a bunch of different other elements including the ability to listen to your gut feeling. No one knows the future but what we do know is history. We have a history of decades of movie data collected by IMDB, TMDB, Rotten Tomatoes, Box Office Mojo. Humans never really changed, sure we do change in our outward appearance over a long course of time but our key characteristics and behavior are hardwired in our DNAs and this will take a large amount of self-awareness to decode. and let’s be realistic no one starts thinking introspectively when deciding on a movie to watch. It’s more of an impulse decision, like when I decided to watch the movie, Dangerous Lies, at 11:00 PM when I was definitely self-aware that I need to be up at 6:00 AM. What I’m trying to say here is if there was one place where data science could be applicable it will be here, in the movie industry. We have large amounts of data from viewers behaviors which we can use to predict their future behaviors.

Take Netflix for instance, I never went to the Netflix search box and typed in Dangerous Lies, it just appeared on my homepage as a recommendation so I just impulsively clicked on it. Regardless, it came of no surprise to me that I actually enjoyed watching it a lot. Netflix has thousands if not millions of movies and TV shows and in order to help customers sieve through their ever-changing catalog, they recommend movies and TV shows to watch based on a viewer watching habit. Netflix understands they can’t offer up their entire catalog to customers so they make use of algorithms which are computer- controlled database sets of rules. Netflix even said publicly that the Netflix experience is controlled by a number of machine learning algorithms such as personalized ranking, search similarity, watch history, ratings, and more. They understand they must curate what to present customers based on what the algorithm feels the viewer would like if not one could spend minutes if not hours searching through the database for the perfect movie.

Netflix even goes as far as tailoring the cover art off movies to fit their viewers taste. According to internal studies by Netflix, they found out that viewers spend an average of 1.8 seconds considering a title before moving to the next and they believe they only have 90 seconds to capture attention before you moved to another activity. They use Aesthetic Visual Analysis, AVA, to capture the perfect personalized thumbnail for each customer. AVA uses tools and algorithms that search Netflix videos for the best images to capture in order to make thumbnails out of them. This way they present to you, after multiple tests, the thumbnail they feel will grab your attention the most.

Because there are so many human decisions, Netflix understands its extremely useful to tag movie titles with very rich information so that customers can gain more insight into a movie, without reading much of the description. Netflix uses a very intricate tagging system to form these tags, there’s even a job at Netflix called a tagger, these people watch every single movie and TV show on Netflix and tag them consistently so that Netflix can have consistent ways to do things anywhere where there’s the human to machine interface such as members trying to look at their homepage and be able to see organized titles with labels in the rows. The tags also help Netflix predict the potential popularity of a title that hasn’t launched yet on the platform. They use this to plot a graph showing the predicted popularity of a tittle against time. They split the time axis into four segments: pitch phase, development & production phase, pre-Launch phase, and launch phase. They use this graph to make decisions at the end of each of the four segments of time about if a title has the green light or if they’ll be pulling the plug on it.

Even Legendary Entertainment, known for King Kong, Dark Knight, etc, uses their own type of AI to predict what type of impression to show to prospective customers in order to lure them into either watching a trailer or buying a movie ticket. If you notice here, the successful companies in the movie industry maintain this success with an eagerness to understand people, and how best can you understand people? Data. It is no more show me your friend and I’ll tell you who you are. Rather, show me your data and I’ll tell you who you are, and even maybe, who you will be. They use the data of people they have for analysis in order to create an audience and make individual predictions from it.

They use this insight to hone in on who they think their ideal audience would be. Using this they create three categories of audiences, the givens, who are avid long-time fans of the movie and will watch it no matter, for example, me for James Bond, the Nevers, who are people that have never watched it and have no intention on watching it, for example, me for Keeping Up With The Kardashians, and in the middle, they have the Persuadables, who are people that are convincible with the right impression at the right moment. Most of their targeting ads and impressions will be geared towards the Persuadables, they have no interest pursuing the Nevers and there’s no use of selling something that is already sold so there’s no need to focus on the givens. This is why most of their marketing in time and ads are spent trying to convince the Persuadables. They take this Persuadables and scored them individually from 0, meaning unlikely, to 100, meaning very likely. They then take a small group and carry out various tests using various marketing campaign strategies targeted to this small group. They then use the successful marketing campaigns and apply them to a broader group of the Persuadables, then as the release date approaches, they downsize this group to make for a more targeted audience.

Reference

78 views

Recent Posts

See All

Comments


Post: Blog2_Post
bottom of page