Core Concepts
Clustering
How narratives are grouped and identified
Semantic Clustering
Semeion uses embeddings to cluster signals by semantic similarity. This helps identify coordinated narratives even when the exact wording differs.
How It Works
- Embedding Generation - Each signal's content is converted to a vector using an embedding model
- Similarity Search - The vector is compared against existing cluster centroids
- Assignment - If similarity exceeds the threshold (default 0.75), the signal joins the cluster
- New Cluster - Otherwise, a new cluster is created with this signal as the seed
Cluster Labels
Clusters are automatically labeled using AI summarization. The label captures the core narrative or theme shared by signals in the cluster.