How design influences capturing ML signals
Machine learning is a method to make predictions based on some set of previous data points. Feed systems are dependent on ML models, which heavily rely on getting feedback from users to learn what is most relevent and engaging content for them. Serendipitous discovery in feed that aligns with personal interests generally creates the highest value and retention.
Design x ML signals principles
Design enables ML to provide engaging content by building interfaces that capture user intent, which are interpreted through signals. The following principles were established to guide the type of signals design needed to capture for the ML models.
1 – Explicitness
How clearly the user's intent is captured through an action
Explicit signals, like selecting an action, give more confidence for a model to make the right decision, while implicit signals, like analyzing dwell time on content, provides more volume.
2 – Frequency
How often a user repeats an action that is providing a signal
3 – Attributability
How accurately an action can be attributed to a signal
The more explicit the action, the more accurately it can be attributed to a signal.