Wherein Sovay, discontent with memory (data) and imagination (theory), turns to her immediate senses to perform feats of multi-dimensional communication.
Do You See What I See? (NLP and Computer Vision)
As Sovay continues her exploration and analysis around streaming media (OTT), she gleans that what is unique about OTT is how it delivers content directly to the viewer. We can know exactly what they were seeing, hearing, and reading at every interaction, what was said just before they hit “back” or what actor was on screen when they paused.
Clearly, Sovay ascertains, this can take us deep into the user experience.
Sovay also knows the biggest advances in modern AI have been in Computer Vision and Natural Language Processing (NLP). Speech can be turned into text and with modern NLP applied, it is possible to detect sentiment, sarcasm, and even proper names. It’s the next best thing to being in the user’s head.
But it’s important to manage expectations around these capabilities. Even when humans are speaking to each other, we are still constantly asking each other for clarification. One limitation of AI is not having enough of a mental framework yet to realize it didn’t understand something correctly.
Norbert Wiener showed in 1948 that all AI systems are limited by the information provided to them. We are just beginning to understand the vast divide between human knowledge and computer knowledge.
Social Listening (Graphs and Shapley Values)
OTT’s unique ability to uncover viewer insights can help us discern when they share, like, or retweet content. We are able to learn from their inbound searches as well. It’s rumored this is how Netflix cast Robyn Wright in House of Cards after uncovering a large number of searches for what she had done after Princess Bride.
These user interactions create two kinds of graphs: One for the content (e.g., when one viewer shares Jason Bateman content, do they also share Laura Linney or Michael Cera content?) and one for the user’s social network (e.g., who do the influencers influence?) These graphs can be loaded (separately) into systems to detect communities, clusters and degrees of separation. This gives us insight into what characters, actors or users move in chorus.
These clusters of users and insights can drive revenue, engagement, and subscriptions. The next question is:
Which members of this group are the real drivers?
A common technique is to use linear regression and see which has a correlation, but this misses critical network effects. Lloyd Shapley tackled this problem in 1951 when he derived Shapley Values. He later earned the Noble Prize in Economics for this method. The idea is simple:
Which players contribute more or less to the team?
The calculations may look intimidating, but they are easily implemented after the analyst has decided what “value” means. This is one of several techniques that have a proven record in resolving complicated market problems, but it is seldom used outside of academia.
As Sovay continues her exploration, she’ll encounter strange new worlds of tigers, content, and game theory. Stay tuned for the next installment of Sovay’s OTT adventure.