Clinical Design Entity Recomendation

Minimal proof-of-concept for:

• multi-signal study similarity (components A–E), and
• similarity-driven entity recommendations (top-K, missing entities, co-occurrence, rerank hooks).

Similarity and recommendations assume clinical documents are already digitized into a knowledge graph (sections/chunks exist as nodes). Entity nodes are then added on top of those source sections, and standardized into canonical ids for cross-study comparison.

Notebooks

• notebooks/study_similarity_components.ipynb — step through A–E, then the full weighted score.
• notebooks/recommendations.ipynb — top-K, missing entities, co-occurrence filtering, reranking.

Extension

Implement ProtocolGraphSource (and optionally EntityStandardizationPipeline) against your property graph and wire real matrices into StudySimilarityInput / the recommendation functions.