Designing Recommender Systems - 8.2
Through the lens of experts and synthesis
This piece continues on from our April newsletter on taste and recommender systems.
So, how do great recommendations come about? By becoming an expert.
And this is where I think through the lens of synthesis.
Synthesis is the construction of a new conceptual whole from a variety of points. It comes from a deep understanding of a field, which is then turned into a new theory or shared concrete understanding. Simply, it's the process of generating new knowledge.
For synthesis to occur, four things are necessary:
Context — In the context of a recommender, this is the recsplanation of why.
Compression — A summarized blurb, not a manuscript.
Multiplicity — Many attempts to learn and train, all naturally a touch different.
Composability — The mid-level representations we can deconstruct. Musically this may be genre, tempo, loudness, etc.
Many people also synthesize best in collaboration with others, typically in the form of conversation. (I am one of these people!)
I’m simplifying a bit here, but these are the core elements we can apply in our design.
If we have these components, we can decompose them to understand what we truly love. I won’t lie to myself, though. It’s not as easy as just piecing parts together. We know that emergence exists! The parts interact differently with each other and become something greater as a whole.
At this point, the path forward should be clear.
A recommender engine is a machine with an aim to suggest something we love.
If the recommender engine were a human, we would call it an expert.
An expert builds their taste through experience, and synthesizes that experience to form knowledge.
Recommender engines synthesize information for users to present them with a suggestion.
But they don’t go far enough. They don’t help us understand why.
Jean Piaget, the famous educational theorist and philosopher, claimed to understand is to invent.
Synthesis is the process of inventing something new from that which we have come to understand.
And thus, the punch. We should design our recommender engines to help us synthesize.
If you’re interested (and a Spotify user), DM me and I’ll share it with you in its early form in the coming weeks.