content discovery with moods
TRANSCRIPT
Content Discovery with Moods Jian Li
Principal Data Scientist, Sky [email protected]
Guardians of the Galaxy Vol. 2
Recommendation engines help customers discover content.
Machine Learning Models
Me & My Devices
What I Watched
The Man Who Knew Infinity
The Trip
True Lies
Recommendations
Step Up
Invincible
White Island
Hot Pursuit
Power On The Box
Little Boy Blue
Recommendation engines help promote new content to customers.
Machine Learning Models
Show on My Devices
Relevant to What I Watched?
The Man Who Knew Infinity
The Trip
True Lies
New Content
Step Up
Invincible
White Island
Hot Pursuit
Power On The Box
Little Boy Blue
Humans have moods. Recommendation machines “talk” in genres. It is difficult for machines to translate moods to genres. So opportunities to promote the most relevant content get missed.
Funny Human Mood
Direct translation with a genre Comedy
Genre
Action Exciting
Hard to translate to one specific genre
Animation Sci-fi Thriller
The labels we choose fruit by are clear and one to one. The labels we choose films can be emotive (mood-based) and “one to many”.
Apple Orange Banana
Funny Adventurous
Exciting
Complete Content Mood Profile
… … … … … …
Our patent pending model learns from content and directly analyses user moods to make ranked recommendations.
Funny 5.2 Adventurous 3.8 Exciting 1.4
Tense Detective mystery
Suspense horror War drama
Ghost thriller …… …… …… …… ……
12.3 10.5 9.2 8.5 … … … …
Exciting Kidnapped
action Secrete services
Buddy cop Military action
…… …… …… …… ……
12.3 9.9 8.4 7.9 … … … … …
Funny Action comedy
Detective comedy Family
animation Spy comedy
…… …… …… …… ……
11.2 9.6 9.5 9.2 … … … … …
Training Data
Machine Learning
Learning semantic relationship between moods and key words
User’s “Momentary” Mood Profile
Semantic Representation of Moods
Funny 60% Adventurous 30% Exciting 10%