Facets: Constraints or Preferences?
Source: https://queryunderstanding.com/facets-constraints-or-preferences-7c25d3a0f4b Author: Daniel Tunkelang
Summary
Should a facet selection be treated as a hard filter (constraint — zero results outside) or a soft preference (boost results matching the facet, but don’t eliminate others)?
Hard Constraints
- Boolean filter: only show items matching the selected facet value
- Problem: can create zero-result pages when the intersection is empty
- User expectation: “I want ONLY brand X, size Y”
Soft Preferences
- Boost items matching facet selection in ranking, but don’t hard-filter
- Allows graceful degradation: show best matches first, then near-misses
- Problem: may confuse users who expect strict filtering
The UX Trust Problem
If users believe facets are hard constraints (as most e-commerce facets behave), and the system softens them silently, it erodes trust. Users will see items that don’t match and assume the search is broken.
ML Approach: Inferring Flexibility
Train a model to predict whether a given user + query + facet combination signals a hard constraint vs. soft preference:
- Features: query type, facet dimension, user history, result set size
- Signal: do users click non-matching items after selecting the facet?
Design Implications
- For navigational queries: hard constraints are appropriate
- For exploratory queries: soft preferences improve discovery
- Hybrid: hard-filter but show “did you mean to include X?” suggestions
Key Concepts
- Hard constraint — boolean filter eliminating non-matching items
- Soft preference — ranking boost for matching items
- UX trust — user mental model of facet behavior
- Flexibility inference — ML model to predict constraint vs. preference intent