Derives a site quality signal from user dwell durations measured against the typical durations for the categories a site belongs to, flagging sites whose users spend less time than category norms predict.
Patent Overview
- Inventor
- Navneet Panda
- Assignee
- Google LLC
- Filed
- 2012-09-28
- Granted
- 2015-10-27
- Application Number
- US 13/631,514
The Challenge
Raw Dwell Times Are Misleading Without Category Context
Different content categories have different typical user dwell durations. A news article naturally produces shorter visits than a reference page; a quick-answer page naturally produces shorter visits than a long-form tutorial. Comparing raw dwell times across sites without category context produces meaningless quality signals. The system needs to read dwell against category-specific baselines, flagging sites that underperform their category norms.
- Categories Have Different Natural Durations — News, reference, shopping, social, entertainment — each category has a different typical dwell. Comparing across categories is comparing apples to oranges.
- Raw Dwell Is Noisy — User attention varies, content length varies, intent varies. A single visit's duration is too noisy to drive ranking. Aggregating across visits and normalizing by category produces a stable signal.
- Underperformance Within Category Is The Signal — A site that consistently produces shorter visits than category-typical for its content type is signaling lower quality. The within-category comparison is what makes the signal interpretable.
- Need Per-Site Category Membership — Each site belongs to one or more categories. The duration measurement happens per category so a site that spans multiple categories gets evaluated against each category's norms.
Innovation
Category Duration Scores Per Website
For each website, the system obtains duration measurements of user device session visits to the site's resources. It obtains data describing the categories the website belongs to. For each category, a category duration score is computed proportional to the duration measurements of the site within that category. The resulting site-by-category duration profile flags sites whose duration scores fall below category norms.
- Collect Duration Measurements — For each website, gather duration data from user device session visits. Each visit produces a measurement of time spent on the site's resources.
- Identify Categories Per Site — Determine the categories the site belongs to. Sites may belong to multiple categories; each gets independent treatment.
- Compute Per-Category Duration Score — For each (site, category) pair, compute a duration score from the visit measurements in that category. The score is proportional to typical visit duration within the category.
- Compare Against Category Norms — For each category, compute the typical duration score across all sites in that category. The site's score is then compared to the norm to identify underperformers.
- Flag Underperforming Sites — Sites whose per-category duration scores fall below category norms are flagged. The flag feeds into the broader site quality signal.
- Apply To Ranking — The quality signal modulates ranking. Sites that consistently underperform category norms on duration get downward adjustments.
Dwell Against Category Norms
The patent's contribution is normalizing dwell by category. Raw dwell is noise; dwell relative to category-typical is signal. The normalization makes user-attention data usable for cross-site quality ranking.
Normalize Before Comparing
Different categories have different natural dwell distributions. Comparing site-A-dwell to site-B-dwell only works when both are measured against their category baselines.
- Duration Measurements — Per-visit time on site, aggregated across visits.
- Category Categorization — Which categories the site belongs to. Multiple categories possible.
- Per-Category Score — Duration aggregate within each category, normalized for cross-site comparison.
Technical Foundation
Inputs And Outputs
Three inputs combine into the per-category duration profile.
- Visit Duration Measurements — Per-session time on site, gathered from user device session data.
- Category Membership — Which categories the site belongs to. From classification or taxonomy data.
- Category Duration Score — Per (site, category) duration aggregate. The metric that drives quality assessment.
Key Insight: Treating dwell as a quality signal requires category normalization. Without it, sites in inherently short-dwell categories (news, quick reference) look bad and sites in inherently long-dwell categories (tutorials, reference) look good, regardless of actual quality. The category normalization is what makes the signal honest.
<\/section>What This Means for SEO
What This Means for SEO
Category-normalized dwell signals shape how dwell-based quality contributes to ranking. Understanding the normalization changes how to think about engagement targets.
- Engagement Targets Are Category-Relative — Your target dwell should match category norms for your content type. A news site optimizing for 10-minute dwells is fighting the category; a tutorial site producing 30-second visits is signaling poor quality within its category.
- Below-Category-Norm Dwell Is Quality Risk — If your dwell consistently lands below the typical for your category, the system reads it as a quality flag. Improve content fit, depth, or formatting to meet category expectations.
- Multi-Category Sites Need Per-Section Treatment — A site with news plus reference plus product content has different dwell norms per section. Optimizing globally for a single dwell target across all sections hurts the ones that don't fit.
- Content Length Should Match Category — Quick-answer categories reward concise content; deep-tutorial categories reward longer content. Length should be calibrated to the dwell norm of the category, not to a global ideal.