2026-02-15
How AFYN Uses Sentence Transformers for Affinity
AFYN needs to compare meaning, not just keywords. Sentence transformers help encode open-ended responses into vectors.
High-level flow
- user submits CoreQuest response
- response text is embedded into vector space
- vectors are compared for semantic proximity
- results are aggregated into KinType and alignment signals
Why embeddings are useful
Embeddings capture context and intent better than keyword overlap.
That allows AFYN to recognize similarity even when two users use different wording.
Output layer
Raw vector similarity alone is not enough. AFYN combines it with model structure so users get interpretable outcomes.