Analysis · Agent 04
Groups near-identical posts and comments so that the same message pushed by many accounts is visible as one coordinated push rather than as unrelated items.
Simplified clustering — not a full ULTRA integration
Items are compared one at a time against existing cluster centroids using multilingual sentence embeddings and cosine similarity. A campaign flag is a lead for an officer to verify, never a finding on its own.
Distinct message groups found.
Reached 3+ near-identical members.
Members in the biggest group.
Total items assigned to a cluster.
Each flagged item is converted to a multilingual sentence vector (MiniLM), so wording differences don't hide a repeated message.
The vector is compared against every existing cluster centre. Above 0.8 cosine similarity it joins that cluster.
If nothing is close enough, it starts a new cluster. The cluster centre is then re-averaged to include the new item.
Once a cluster reaches 3 members it is marked a possible campaign and surfaced to an officer.