Total Mentions
-
Unique Entities
-
Sources
-
Avg Daily
-
Daily Mentions
By Source
Stance Distribution
Top Entities
Entities with accelerating mentions (velocity >= 2x previous period)
Total Mentions
-
Dominant Stance
-
Mention Types
Stance Balance
Stance by Source
Most Discussed Entities
Stance by Entity Type
Recent Viewpoints
Top Narratives
Topic Cloud
Narrative Clusters
-
Top Influence
-
Last Computed
-
Methodology
Influence Score (0-100)
Each mention gets a score based on its source's engagement metrics, log-normalized and scaled to 0-100 within each source type:
| Source | Metrics | Weights |
|---|---|---|
| Tweet | views, likes, retweets, replies | 0.4, 0.3, 0.2, 0.1 |
| HackerNews | score, comments | 0.6, 0.4 |
| GitHub | stars, forks, today_stars | 0.5, 0.2, 0.3 |
| ProductHunt | votes | 1.0 |
| Podcast | views, likes | 0.6, 0.4 |
| Crunchbase | funding, investors, rank | 0.5, 0.3, 0.2 |
| Newsletter | (no metrics) | fixed 30 |
Formula: score = log(1 + metric) * weight, then normalized to 0-100 per source type.
Narrative Clustering
All mentions are clustered once (full dataset) by semantic similarity using embedding vectors (Gemini Embedding 2, 3072d). Time windows filter dynamically at query time:
- Load all mentions with embeddings + influence_score > 0 (~800k)
- MiniBatchKMeans clustering (K=2000, O(N) memory)
- Filter clusters with < 3 members
- LLM (Claude Haiku 4.5) summarizes each cluster into a label + summary
- Time window filters cluster members by observed_at, re-aggregating stats dynamically
Ranking: Clusters are ranked by total influence score within the selected time window (sum of member scores in that period), not by mention count.