September 25, 2017
It’s safe to say the market for media streaming services has grown increasingly competitive in the last five years. Each platform differentiates itself by its content or features, but few have succeeded with automated recommendations. Customer retention is key, and platforms like Spotify and Netflix use artificial intelligence for good (efficiency) rather than evil (total domination), despite what the technophobes say.
With industry giants like Netflix, Spotify and Amazon optimizing deep learning and collaborative filtering, automated recommendations have become the present and future of media discovery. Artificial intelligence gets a bad rap in movies like Terminator, where AI destroys jobs and renders mankind irrelevant. In reality, AI creates new jobs. Since AI is neither autonomous nor sentient, there will always need to be people managing these technologies. AI is doing our dirty work: repetitive, menial tasks that cut into the 20% of non-routine tasks that drive 80% of value creation.
Not only will our work be more productive, but also our leisure time will be more efficient. Our brains have evolved with a limited working memory, meaning we can only consider a small set of information when making a decision. These days, more time is spent looking for content to watch than ever before. Based on an Ericsson study, US consumers spend 45% more time choosing what to watch on platforms like HBO GO, Netflix and Amazon than they do with scheduled TV. However, 63% of video on demand consumers are very satisfied with the content, compared to 51% very satisfied scheduled TV consumers.
Despite our eventual satisfaction, the more options we have, the more likely we are to be stressed about making decisions. This stress-inducing “choice paralysis” becomes ever-increasing due to the of video-on-demand platforms. The dissatisfaction of having to make choices is also demonstrated online and speed dating. When confronted with more options, people tend to make no decision at all and walk away feeling frustrated. People make decisions by using heuristics, basically ignoring some information and focusing on only one aspect of the problem.
How often have you started watching something on Netflix and within the first five minutes switched to something else? AI can help us make decisions and cut down on the unnecessary stress we feel when choosing what to watch for an hour of downtime. These technologies have been designed to streamline menial tasks and narrow down choices that are psychologically overwhelming. This allows humans to free up mental capacity while optimizing complex, nuanced and emotional decisions.
For companies like Spotify (140 million users) and Netflix (103.95 million subscribers), recommending content isn’t exactly a cakewalk. Both companies use collaborative filtering to determine preferences from historical usage data. Rather than comparing similar content, collaborative filtering compares users with similar tastes in order to look beyond basic genres of music and video.
Netflix creates suggestions based on preferences for actors, directors, plotlines, and similar characters in order to improve recommendations, which can help in situations where traditional comparison data is sparse or unavailable. For instance, with the launch of the famed show Stranger Things, 20% of fans had never watched the horror genre before. By breaking down its 103.95 million subscribers into 1,300 taste communities, Netflix has been able to easily filter recommendations based on affinities present in each of these taste communities.
Spotify takes advantage of affinity data in similar ways, by creating profiles of each user’s individualized taste in music, and grouping that information into clusters of artists and micro-genres. This affinity information is pulled from over two billion user-generated and professionally curated playlists, and is thus synthesized into the custom “Discover Weekly” playlists. Spotify pulls data from artists, albums, lyrics, text web reviews, interviews and an audio signal itself. Many labels and artists have requested to seed Discover Weekly playlists with their content, yet Spotify refuses inorganic content on the playlists.
One shortfall of the Discover Weekly playlists is the challenge of finding obscure or new music, especially if users aren’t aware of the group. This phenomenon is called the cold-start problem, which happens when a user joins and has yet to listen to much music, leaving few (if any) consumption patterns to analyze. This is combated with the convolutional neural network, which reviews and analyzes the acoustics of a song to discover other tracks that have similar patterns. It is also possible to source recommendations in deep content and learning. A regression model was trained to predict the latent representations of songs obtained from the collaborative filtering model.
Spotify’s previous product manager, Matt Ogle, stated that they have the technology to “ensure that if you’re the smallest, strangest musician in the world, and doing something that only 20 people in the world will dig, we can now find those 20 people and connect the dots between the artist and listeners.” Suggestions from Discover Weekly tend to be left-of-field rather than vanilla in order to keep listeners on their toes. Outlier detection is notoriously used in financial security to detect fraudulent charges, but can also be used in recommendation engines. The detection determines particular usage that’s part of normal pattern behavior. This preventative measure ensures that when your account is shared, the abnormal behavior will not be incorporated in your Discover Weekly playlist. If you’re a fan of heavy metal music, but have to DJ at your grandmother’s birthday party, your playlist on Monday won’t be flooded with Elvis or Big Band jams.
The Discover Weekly playlists have been so successful based on a simple behavior psychology theory. Spotify’s core strategy is for subscribers to view the platform as indispensable. The exciting reward of new music every week with an expiration date motivates users to open the app and tune in every Monday. Complementing the Discover Weekly habit loop is the Friday Release Radar playlist. This playlist showcases new tracks from artists the user already listens to. Friday is a strategic move because it’s the global release day for new music, meaning that Spotify can get an artist’s new tracks onto your radar fresh out of the recording studio.
Automated intelligence will help video on demand platform users make decisions more quickly, rather than wasting time scrolling through endless content. The automated recommendations also help users step outside of their comfort zone and encourage them to broaden their entertainment horizons. Spotify and Netflix will never run out of content to keep current users hooked, and these platforms are already becoming indispensable for younger generations ditching cable TV and radio. The future for content and entertainment is bright, and technophiles are looking forward to watching what happens.
Author: Brooke Bode, Intern at Engage