Recommendation and ranking narrow visible viewpoints
Aliases: recommender diversity · exploration · filter bubble
What it is
A filter bubble is the narrowing of viewpoints actually seen when recommendation and ranking repeatedly show familiar positions based on historical clicks, relations, or similarity. It is not inevitable: it follows relevance optimisation without exploration or diversity constraints.
Why it happens
Clicks and dwell become both preference evidence and future ranking input. Content pushed down loses the chance to be chosen, apparently proving irrelevance. Network and popularity signals concentrate attention. People may not know what failed to appear and cannot correct the range.
Studying it
- Paradigms: compare exposure diversity, cross-view contact, and reading before/after recommendation; run ranking experiments.
- Variables: candidate pool, rank, exposure, click, diversity, exploration, and control.
- Methodological caution: exposure is not understanding or acceptance; distinguish seeing, reading, and interacting.
Where it stops holding
Forced diversity can insert low-quality or harmful content. Task relevance differs, and safety or expertise cannot be replaced by viewpoint balance.
Applying it
- Reserve explainable exploration positions and diverse sources with user-controlled range.
- Show why items are recommended and uncovered topics.
- Audit exposure and quality across viewpoints and new sources.
- Verification: compare source range, satisfaction, misinformation, and harm under default and diversity settings.
Related
- Same group: V7.07.1 Discussion in homogeneous groups makes positions more extreme · V7.07.3 Dissenters fall silent in homogeneous settings · V7.07.4 Simple exposure to opposing views can deepen division · V7.07.5 Permeable group boundaries set the ceiling on polarization
- Nearby: V8.03 Content ranking and visibility · V9.07 Group judgment and conformity
- Search terms:
filter bubble·recommender diversity·exploration