• The classic RecSys trap: a brand-new user or item has no interaction history, so behavior-based methods have nothing to work with.
• Hits Collaborative Filtering hardest (it uses only interactions).

Two cases

• New user (no history): fall back to
  • popularity (show what is popular overall),
  • context (location, device, time),
  • a few onboarding questions to seed a profile.
• New item (no interactions): use content features.
  • The item tower of a Two-Tower Model produces a vector from features even with zero clicks, so two-tower / content-based handles new items. Pure collaborative filtering cannot.

One-line answer

• Cold start = missing interactions. The fix is to lean on features and context until interactions accumulate.

References