Update skill_extraction.py

skill_extraction.py

@@ -1,3 +1,5 @@
+# skill_extraction.py
+
 import json
 import spacy
 import numpy as np
@@ -18,9 +20,12 @@ model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
 with open("skills_vocab.json", "r", encoding="utf-8") as f:
     SKILL_VOCAB = json.load(f)["skills"]
 
-# Precompute skill embeddings
+# Precompute embeddings
 skill_embeddings = model.encode(SKILL_VOCAB, convert_to_numpy=True, normalize_embeddings=True)
 
+# Map skill -> embedding for quick lookup
+SKILL_TO_EMB = {skill: emb for skill, emb in zip(SKILL_VOCAB, skill_embeddings)}
+
 def extract_skills(text, threshold=0.50):
     doc = nlp(text)
     sentences = [sent.text.strip() for sent in doc.sents if sent.text.strip()]
@@ -29,17 +34,17 @@ def extract_skills(text, threshold=0.50):
         return []
 
     sentence_embeddings = model.encode(sentences, convert_to_numpy=True, normalize_embeddings=True)
+    skill_confidences = {}
 
-    extracted = []
-
-    # Iterate skill-by-skill
     for j, skill_emb in enumerate(skill_embeddings):
         sims = np.dot(sentence_embeddings, skill_emb)
         max_sim = float(np.max(sims))
         if max_sim >= threshold:
-            best_sentence = sentences[int(np.argmax(sims))]
-            extracted.append((SKILL_VOCAB[j], round(max_sim, 2), best_sentence))
+            skill_confidences[SKILL_VOCAB[j]] = round(max_sim, 2)
 
-    # Sort highest confidence first
-    return sorted(extracted, key=lambda x: x[1], reverse=True)
+    return sorted(skill_confidences.items(), key=lambda x: x[1], reverse=True)
 
+# --- NEW FUNCTION ---
+def get_skill_embedding(skill_name):
+    """Return the embedding vector for a skill, or None if not in vocab"""
+    return SKILL_TO_EMB.get(skill_name)