What You'll See in the Dashboard
Confusion and Emotional State scores appear in the dashboard alongside ClickStream's other behavioral scores. High Confusion scores highlight pages where visitors struggle to find information — a direct UX improvement signal. Emotional State provides an inferred sentiment label for each visitor session.
Business Actions: Alert your UX team when Confusion > 60 on key landing pages. Use Emotional State trends to decide where supportive messaging will help visitors showing negative sentiment.
Model 6: Confusion Detection
The confusion score measures how much difficulty a user is having finding information or completing tasks on your site. While related to frustration, confusion is distinct: a frustrated user knows what they want but cannot get it; a confused user does not know where to look or what to do next.
The 7 Confusion Signals
The confusion model combines behavioral signals like the following. The weights shown are illustrative — signal weighting is tunable per site.
| Signal | Weight | Detection Method |
|---|---|---|
| Circular navigation | 0.22 | Visiting the same 3+ pages in a loop within 60 seconds |
| Excessive menu scanning | 0.18 | Mouse hovering over multiple nav items without clicking (>3 items in 5s) |
| Search-after-navigation | 0.16 | Using site search immediately after navigating to 2+ pages |
| Long dwell with no interaction | 0.14 | Page visible for 30+ seconds with zero mouse/keyboard activity |
| Rapid page switching | 0.12 | Visiting 3+ different pages within 15 seconds (not linear flow) |
| Scroll-to-top-and-restart | 0.10 | Scrolling 60%+ down then returning to top and starting over |
| Help/FAQ seeking | 0.08 | Navigating to help, FAQ, or contact pages mid-task |
The 4 Confusion Types
Not all confusion is the same. ClickStream classifies confusion into four types that suggest different remediation strategies:
| Type | Pattern | Typical Cause | Remediation |
|---|---|---|---|
| Lost | Circular navigation + menu scanning | Poor information architecture, unclear labels | Breadcrumbs, improved nav labels, contextual links |
| Searching | Search-after-navigation + help seeking | Cannot find specific information or action | Better search, inline guidance, tooltips |
| Stuck | Long dwell + scroll restart, no forward progress | Unclear how to complete the current task | Step indicators, progressive disclosure |
| Overwhelmed | Rapid switching between similar pages/options | Too many options without clear differentiation | Comparison tools, recommendations, filters |
Confusion Signal Shape (conceptual)
Finding Confusion Hotspots
Confusion scores are persisted per event to ClickStream's analytics store (the clickstream_scores dataset), so the dashboard can rank pages by confusion frequency and severity over any date range — no warehouse queries required.
Model 7: Emotional State Classification
The emotional state model infers a user's emotional valence from behavioral micro-signals -- primarily mouse dynamics and typing cadence. This is one of ClickStream's most nuanced models, and we approach it with careful attention to ethical boundaries.
The 8 Emotional States
| State | Behavioral Indicators | Business Implication |
|---|---|---|
| Curious | Varied pace, diverse page types, exploratory scrolling | User is in discovery mode. Suggest related content. |
| Engaged | Smooth mouse movement, steady scroll, consistent click intervals | User is on-task. Do not interrupt. |
| Frustrated | Rage clicks, cursor thrashing, rapid back-navigation | Immediate intervention needed. See Part 1. |
| Confused | Circular navigation, menu scanning, scroll restarts | Pair with the confusion score to locate the UX gap. |
| Excited | Fast but purposeful movement, quick clicks, deep scroll | High receptivity. Surface the primary CTA. |
| Decisive | Direct paths to targets, short click preparation, minimal backtracking | User is ready to act. Remove friction from the path. |
| Hesitant | Slow mouse, long pauses before clicks, hover-without-click | User needs reassurance. Show social proof. |
| Neutral | No dominant behavioral pattern | Baseline state. No action needed. |
Mouse Dynamics Signals
Mouse movement patterns are rich behavioral signals that reflect cognitive and emotional state. ClickStream's model draws on signals like the following from mouse event streams:
- Velocity profile: Average speed (px/s) and velocity variance. Smooth, consistent velocity suggests focus; erratic velocity suggests agitation.
- Curvature index: Ratio of actual mouse path length to straight-line distance between start and end points. High curvature indicates hesitation or exploration.
- Pause frequency: Number of mouse stops (<2px movement for 500ms+) per minute. Frequent pauses suggest reading or deliberation.
- Click preparation time: Time between mouse arrival at a target element and the click event. Longer preparation = more deliberation.
- Direction entropy: Shannon entropy of mouse direction changes. Low entropy = purposeful movement; high entropy = scanning or confusion.
Typing Cadence Signals
For pages with form inputs, typing patterns provide additional emotional indicators (note: ClickStream's scoring uses interaction timing, not form contents):
- Keystroke interval variance: Consistent typing speed suggests confidence; variable speed suggests uncertainty or correction.
- Backspace ratio: Percentage of keystrokes that are deletions. High ratios indicate self-correction and potential frustration.
- Inter-field pause: Time between completing one field and starting the next. Long pauses suggest the user is thinking or confused about what to enter.
- Typing burst patterns: Fast-pause-fast patterns suggest copying from another source; steady typing suggests known information.
Confidence Scoring
Every emotional state classification comes with a confidence score (0.0–1.0). When no state can be inferred with sufficient confidence, the classification is reported as neutral.
Ethical Considerations
Emotional state inference from behavioral signals raises important ethical questions. ClickStream's approach is governed by these principles:
- No individual emotion tracking: Emotional states are used for aggregate UX analysis and real-time experience optimization, never for individual profiling or targeting based on emotional vulnerability.
- Opt-out available: Visitors can opt out of behavioral analysis entirely through the privacy controls embedded in ClickStream's SDK.
- No manipulation: Emotional state data must not be used to exploit users in vulnerable states (e.g., showing high-pressure sales tactics to frustrated users). ClickStream's recommended actions are focused on helping users, not manipulating them.
- Transparency: Sites using ClickStream's emotional state model should disclose behavioral analytics in their privacy policy.
- Data minimization: Analytics storage retains the computed emotional classification and confidence score, not raw input streams.
- No cross-site emotional profiling: Emotional states are session-scoped and never aggregated into a cross-site emotional profile.
Confusion × Emotion Interaction Table
When confusion and emotional state are combined, they reveal nuanced user experience issues:
| Confusion Type | Engaged | Hesitant | Decisive | Frustrated |
|---|---|---|---|---|
| Lost | Methodical search. Improve nav labels. | Lost and unsure. Show breadcrumbs. | Knows the goal, cannot find the page. Add search. | Broken navigation. Critical UX fix. |
| Searching | Willing to dig. Improve search relevance. | Unsure what to look for. Suggest queries. | Specific target in mind. Surface quick links. | Search is failing them. Fix zero-result queries. |
| Stuck | Process is unclear. Add step indicators. | Unsure what to enter. Add field hints. | Wants to finish fast. Reduce steps. | Task is broken. Emergency fix. |
| Overwhelmed | Comparing carefully. Show comparison table. | Too many options. Show recommendations. | Ready to choose. Offer a default selection. | Options are confusing. Simplify pricing. |
Use Cases
UX Audit Automation
Review confusion hotspots weekly to identify the pages and flows that generate the most user confusion. Prioritize UX fixes based on confusion score severity and unique visitor count.
Dynamic Help Systems
When confusion score exceeds 60 and the emotional state is hesitant, dynamically show contextual help tooltips or offer a guided tour for the current page section.
Content Optimization
Pages with high confusion scores and low engagement trajectories are candidates for content restructuring. Use the confusion type to determine whether the issue is information density, language clarity, or missing information.
Checkout Flow Optimization
Stuck-type confusion during checkout is one of the highest-impact UX issues. Combine confusion scoring with form friction analysis (covered in Part 7) for a complete picture of checkout friction.