Can implicit feedback reveal both preference and confidence?
When users take implicit actions like purchases or watches, do those signals carry two separable pieces of information: what they prefer and how certain we should be? Explicit ratings can't make that distinction.
When users give explicit star ratings, the number expresses preference directly: a 5 means strong like, a 1 means strong dislike. But most real systems must work from implicit signals — purchases, watches, clicks, browsing time — where the user never declares a preference. The conventional move is to threshold the implicit signal and pretend it is explicit feedback.
Hu, Koren, and Volinsky argue this is the wrong reduction. Implicit observations carry two distinct magnitudes that should be modeled separately. The first is a preference estimate: did the user like this item or not? The second is a confidence level: how sure are we? A user who watched a movie three times provides high-confidence positive preference. A user who watched ten seconds and quit provides low-confidence and probably negative preference. A user who never saw the item provides low-confidence with no directional information.
Explicit ratings collapse these two magnitudes into one number. A "3 stars" rating is medium-confidence medium-preference; the data structure can't distinguish "I'm sure I felt mediocre" from "I'm not sure how I felt." Implicit data forces the partition — and once you make it explicit, the resulting model handles the asymmetry between observed and unobserved entries naturally. Unobserved is not "preference zero" but "confidence zero, preference unknown."
Inquiring lines that read this note 12
This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.
How can we distinguish genuine user preferences from measurement artifacts?- How can consistency across measurement conditions identify genuine versus constructed preferences?
- How do implicit signals like clicks capture preference more reliably than explicit ratings?
- Why do explicit ratings fail to capture uncertainty in user preferences?
- How does implicit feedback structure differ from explicit ratings mathematically?
- What consistency tests could distinguish constructed from genuine preferences?
- How do confidence signals differ between implicit feedback and explicit ratings?
Related concepts in this collection 4
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
-
Why do the same users rate items differently each time?
User ratings are assumed to be clean preference signals, but do they actually fluctuate unpredictably? This matters because recommender systems rely on ratings as ground truth, yet temporal inconsistency and individual rating styles may contaminate that signal.
extends: noise in explicit ratings makes the case for confidence-as-separate-channel stronger
-
Can language models bridge the gap between critique and preference?
When users express what they dislike rather than what they want, can LLMs reliably transform those critiques into positive preferences that retrieval systems can actually use?
complements: critique-as-feedback is a third class — natural-language negative signal that transforms into preference with separate certainty
-
Why does multinomial likelihood work better for ranking recommendations?
Explores whether the choice of likelihood function in VAE-based collaborative filtering matters for matching training objectives to ranking evaluation metrics. Why items should compete for probability mass.
complements: multinomial likelihood operationalizes the implicit-feedback insight by ranking items competitively rather than scoring them independently
-
When can greedy bandits skip exploration entirely?
Under what conditions does natural randomness in incoming contexts eliminate the need for active exploration in contextual bandits? This matters for high-stakes domains like medicine where exploration carries real costs.
complements: confidence in implicit feedback is the bandit-style uncertainty signal that exploration policies act on
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Collaborative Filtering for Implicit Feedback Datasets
- I like it... I like it not: Evaluating User Ratings Noise in Recommender Systems
- Neural Collaborative Filtering
- Measuring Human Preferences in RLHF is a Social Science Problem
- Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains
- Multi-Task End-to-End Training Improves Conversational Recommendation
- Fast and Slow Learning From Reviews
- Self-Improving Model Steering
Original note title
implicit feedback resolves into preference and confidence — two paired magnitudes that explicit ratings cannot represent