Have you ever noticed how difficult it is to find something to watch on a streaming service? In this presentation, I examine the reasons for that issue and discuss a solution involving an improved recommendation engine which compares the content itself rather than the viewing habits of users.
8. Consider recommendation engines: They rely on the premise
that similar viewers will watch similar content.
?
Theoretically, if someone watches three movies and a second
person watches two of those same movies, the second person
might also be interested in the third movie.
9. This is faulty inductive reasoning.
It assumes there is a discernable pattern in the video
selections of those two people, and it then attempts to
describe an algorithm to match that pattern.
10. If there were a pattern,
more data should lead to
more accurate predictions.
But what if the underlying
assumption is wrong?
11. If there is no pattern, more data isn’t helpful.
In fact, there are so many variables that go into choosing
something to watch that it’s functionally random.
12. Even if you could find some
extraordinarily complicated
algorithm to explain the last 50 years
of winning numbers, you still
wouldn’t be able to pick tomorrow’s
numbers.
It’s like picking winning lottery
numbers.
17. For example, if you searched
for a “ghost” movie, the
results would include a wide
range of films with little in
common, and you would still
have to page through the
results to find what you
want.
18. A search for an actor or
director will produce similar
results. Michael Caine was in
all of these films, but they
have little else in common.
21. Viewers most
often want to
find a film
that is similar
to something
they’ve
enjoyed in the
past.
22. It’s roller coaster
psychology.
After a terrific roller
coaster ride, you want
to either get back on
or find an even better
roller coaster. After
experiencing a great
film, you either want
to re-watch it or find
a similar one that’s
even better.
23. This is why movie franchises and
sequels are so popular.
24. Suppose you wanted to see a
movie like Cinderella (1950).
Which of these films would you choose?
Citizen Kane (1941) Dirty Harry (1972) Sleeping Beauty (1959) Risky Business (1983) Cleopatra (1963) Mean Girls (2004)
25. The answer is obvious.
But could a computer be
taught to watch movies
and decide how similar
they are?
And could that computer
make less obvious
connections among
films?
26. The answer is YES! Using machine
learning, and analytics, computers can be
trained to watch motion pictures and
turn their observations into measurable
data.
27. Computers can observe the basic principles of aesthetics (such
as color, line, and movement), as well as look for film techniques
(like aspect ratio, number and types of edits, interior vs. exterior
shots, and facial recognition).
28. Computers can also listen to sounds (such as music, crowd noise,
vehicles, weapons, animals, and even moments of silence).
29. Computers can detect narrative elements too, (such as dialogue,
mood, theme, or the three-act structure).
30. Then, when the data is captured and analyzed, it can be
combined with other known data (such as release date, cast and
crew, production studio, budget, or writers). Ultimately, when all
of the data has been analyzed, the computer will be able to
conclude how similar two films are.
31. 90% 5%
For example, it might suggest that The Maltese Falcon
(1941) is 90% similar to The Big Sleep (1946) but only 5%
similar to Alien (1979).
32. On the other hand,
the machine might
also be able to spot
some less obvious
connections.
For example, how
similar are The
Three Musketeers
(1973) and Justice
League (2017)?
33. Over time, the machine will learn which
metrics produce the best results and
how to weigh those metrics against each
other, to streamline the process and
produce the best results. From there, the
database will be able to legitimately recommend films
based on viewing habits, or allow a user to search for
films that are genuinely similar to movies they already
love.
34. I hope you’ve enjoyed this
presentation. Unfortunately,
there is much more to say about
this than I could possibly put on
these pages. Comments and
criticisms are welcome, and if you
would like more details on how
this would work, I’d love to
connect with you.
Contact information:
Alan G. Shimp
3L at UCLA School of Law
Los Angeles, CA
LI: linkedin.com/in/alan-shimp
Thank you for viewing.