S11-Mo. Mobile-App Recommendation · Search Diversity (Momo-Specific)
Published 2026-05-14Updated 2026-06-302 min read
Momo differentiation point ⭐ — Monitors recommendation diversity and long-tail exposure so that the massive SKU asset stays discoverable in the mobile app.
1. URL · Persona
/recommendation-diversity· P3 (Search · Recommendation · MarTech)
2. User Story
P3 — Track recommendation diversity score, new-product exposure ratio, and long-tail SKU exposure rate by category, segment, and time slot.
3. Data Mix
| Data | Source |
|---|---|
| Product (LongTailSKU flag) | Neptune |
| Recommendation · Search exposure logs | OpenSearch |
| OrderTransaction | Neptune |
| User behavior (exploration vs purchase) | CartEvent + SearchEvent |
4. Metrics
| Metric | Definition |
|---|---|
| Diversity score (intra-list) | Category · brand distribution entropy within recommendations |
| New-product exposure ratio | Share of SKUs launched within last 30 days |
| Long-tail exposure rate | Share of SKUs with monthly transactions below SKU average |
| Exploration-purchase balance | Click / Purchase ratio within normal range |
5. Processing Pipeline
1. Recommendation exposure log dataframe
2. Compute category · brand entropy
3. Compute new-product · long-tail ratios
4. Per-category diversity score → auto-alert when below threshold
5. A/B comparison (diversity ↑ vs CTR ↑ tradeoff)
6. Output UI
- Left: Per-category diversity-score bars
- Center: New-product · long-tail exposure time series
- Right: Exploration-purchase balance score
- Bottom: Diversity ↑ A/B results
7. Guardrails
- Recommendation-algorithm business-information protection
- A/B test results shown with statistical confidence intervals
8. Demo Scenarios
- Per-category diversity score → "Summer tent" 0.31 (low — concentrated in 1-2 popular SKUs)
- Long-tail exposure ratio → 8% → 12% after diversity-corrected recommendation simulation
- A/B results → After diversity ↑: CTR -2%, GMV +6% (long-horizon efficiency ↑)