Social Discovery

Social discovery (stage 4, handle_enumeration, plus stage 5 identity_matching) finds social profiles that impersonate a brand. It runs in parallel with record information and fingerprinting.

Inputs

Brand context — the brand name and the brand’s official social handles.
Official handles — twitter, linkedin, instagram, facebook, youtube, tiktok, github.

Generation

The social/ package enumerates candidate handles from the brand name and existing handles:

Handle permutation — variations of the official handles (prefixes, suffixes, separators, numbers).
Search dorking — web searches for pages impersonating the brand, backed by SerpAPI or Google Programmable Search (DORK_BACKEND), producing candidate profile URLs.
Maigret probing — for candidates, availability/ownership is probed with Maigret (social/availability.py), which checks platform endpoints to confirm a profile exists and belongs to a handle.

Identity matching

Stage 5 (social/identity_matching.py, ~433 lines) matches discovered profiles against the brand:

Each candidate profile’s identity is compared with the brand’s official profiles.
Matching uses embedding-based comparison — text (display name, bio) is embedded with a sentence-transformer model (EMBEDDING_MODEL, default sentence-transformers/all-MiniLM-L6-v2) and compared for similarity.
Results are SocialMatchResult objects recording match signals.

Outputs

The phase emits SocialFinding objects with:

Identity — platform, handle, display name, profile URL, avatar path, bio.
Audience — follower/following/post counts, account creation date, verification status.
Detection metadata — detection_method (permutation / dork / probe), match_signals, cross_links.
Scoring — a SocialScoringResult severity score, refined later by the severity assessment stage.

Configuration