Should recommendations cross community boundaries?

As a product engineer, I’m uneasy about recommendation systems carrying behavior from one online community into another without making that transfer obvious. A person following labor-organizing discussions in one space may not want those interests attached to their entertainment, professional, or family-facing feed. Better discovery is real value, but silent context-switching turns ordinary activity into an invisible identity profile.

My default would be separate preference profiles by community, with visible signal provenance when a recommendation crosses that boundary: “shown because you followed X elsewhere.” Add a user-controlled reset and a genuine opt-in for broader sharing. This is more friction and may reduce relevance, but contextual privacy should outweigh marginal engagement gains. Is cross-community personalization useful infrastructure, or unacceptable invisible identity tracking? I’d like to hear technical alternatives or real-world examples.

I’d go further than disclosure. Provenance labels are useful for debugging, but they don’t undo the inference. Make the default model community-scoped, then let users deliberately create a broader profile. Technically, that is a safer boundary than trying to explain a giant blended feature vector after the fact.

Agreed. “Opt out” is a weak control when most people never know what they are opting out of. I’d want a per-community data ledger: what signals exist, where they can flow, and a delete/reset action that actually removes their recommendation influence.

The objection I hear from product teams is that separate profiles make onboarding and discovery worse. That may be true, but it is a product cost, not a reason to hide the trade-off. Ask permission at the moment a user wants cross-community discovery, not buried in account settings.

The DSA is relevant here, but carefully: it requires plain-language explanations of recommendation parameters and user options, and very large platforms must offer a non-profile-based alternative. It does not appear to establish a universal requirement for separate community profiles or explicit opt-in to every cross-community signal.

Visible provenance can itself leak information. If a shared device or public feed says “because you joined a support group,” the explanation becomes a disclosure surface. I’d use coarse categories by default, with detailed provenance available privately in an account view.

I’m less convinced that community boundaries are stable enough to be the primary model. People often join several spaces precisely because they want ideas to travel. The answer may be a user-managed “context bridge” between named communities, rather than a universal ban or platform-wide blending.

The interface matters more than another settings page. A small “from another community” label beside the item, with one-tap dismiss and reset, teaches the norm at the moment it matters. If users repeatedly reject those suggestions, the system should treat that as a boundary, not just a negative click.

If the useful feature needs a consent dialog, a provenance label, three profile types, and a reset console, perhaps the useful feature is not as useful as advertised. Keep communities separate by default. Let enthusiasts connect them manually.