upload scifact_features.py

scifact_features.py

@@ -0,0 +1,34 @@
+"""Embedding feature builders for claim-document relevance classification."""
+
+from __future__ import annotations
+
+import numpy as np
+
+
+def e5_queries(texts: list[str]) -> list[str]:
+    return [f"query: {text}" for text in texts]
+
+
+def e5_passages(texts: list[str]) -> list[str]:
+    return [f"passage: {text}" for text in texts]
+
+
+def pair_features(model, claims: list[str], documents: list[str], show_progress_bar=False):
+    """Build standard sentence-pair features from two embedding vectors.
+
+    q and d alone give the classifier raw semantic position. abs(q-d) exposes
+    distance dimensions. q*d exposes alignment dimensions. cosine gives a
+    single retrieval-style similarity signal.
+    """
+    q = model.encode(
+        e5_queries(claims),
+        normalize_embeddings=True,
+        show_progress_bar=show_progress_bar,
+    )
+    d = model.encode(
+        e5_passages(documents),
+        normalize_embeddings=True,
+        show_progress_bar=show_progress_bar,
+    )
+    cosine = np.sum(q * d, axis=1, keepdims=True)
+    return np.hstack([q, d, np.abs(q - d), q * d, cosine])