Detect Trends
Before They Go Viral
Advanced community detection on social graphs. Early identification of emerging Instagram trends to empower creators with actionable, data-driven insights.
Trend Analysis Report
This demo presents a search for visual micro-trends in a community of cat memes channels.
š„ = Positive predicted uplift | š = Negative predicted uplift
Seed Analysis
@alsu_leyman
Niche: Photographer
Winning Theme: Intimate Editorial & Soft Tones
High-fashion editorial focusing on intimacy. Key elements include soft lighting, warm/neutral color palettes, and skin-focused compositions.
Significant over-performer: +231% Views on average. This style is the strongest driver of engagement in the niche.
+136% Views 
+80% Views 
+40% Views Performance Risk: Luxury Textures & Classic Portraiture
Traditional luxury fashion elementsāfur, leather, and heavy jewelry. The focus is on tactile materials and "power" posing with clear eye contact.
Significant under-performer: -79% Views on average. The visual complexity and distance from the subject appear to correlate with lower engagement.
-98% Views 
-77% Views 
-59% Views Strategic Recommendations
Do
Shift the visuals toward Soft Tones. Incorporate an intimate, editorial vibe.
Avoid
Description of the features causing negative drift.
Technology
Community Seeding
Input a seed Instagram handle to define your target niche. Our system begins by mapping the specific social graph surrounding your content category.
Feature Embedding Analysis
Every recent post within the niche is decomposed into high-dimensional embeddings. We classify visual featuresāfrom specific accessories to camera anglesāto group content by aesthetic DNA.
Lift Ranking
Using within-entity demeaning and fixed-effects residual analysis, we isolate the marginal performance lift of specific features, prioritizing trends with the highest estimate view uplift.
Predictive Analysis
The pipeline concludes with a high-fidelity uplift projection for the next niche post incorporating the identified feature. Each forecast is synthesized with a formal prediction interval, providing a quantified confidence level for the expected marginal viewership gain.