Update skill_extraction.py

skill_extraction.py

@@ -27,6 +27,10 @@ skill_embeddings = model.encode(SKILL_VOCAB, convert_to_numpy=True, normalize_em
 SKILL_TO_EMB = {skill: emb for skill, emb in zip(SKILL_VOCAB, skill_embeddings)}
 
 def extract_skills(text, threshold=0.50):
+    """
+    Extract skills from text.
+    Returns list of (skill, confidence, evidence_sentences)
+    """
     doc = nlp(text)
     sentences = [sent.text.strip() for sent in doc.sents if sent.text.strip()]
 
@@ -35,14 +39,25 @@ def extract_skills(text, threshold=0.50):
 
     sentence_embeddings = model.encode(sentences, convert_to_numpy=True, normalize_embeddings=True)
     skill_confidences = {}
+    skill_evidence = {}
 
+    # 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))
+        sims = np.dot(sentence_embeddings, skill_emb)  # similarity across sentences
+        max_sim_idx = int(np.argmax(sims))
+        max_sim = float(sims[max_sim_idx])
+
         if max_sim >= threshold:
-            skill_confidences[SKILL_VOCAB[j]] = round(max_sim, 2)
+            skill = SKILL_VOCAB[j]
+            skill_confidences[skill] = round(max_sim, 2)
+            # Take the sentence with highest similarity as evidence
+            skill_evidence[skill] = sentences[max_sim_idx]
+
+    # Return sorted list of (skill, confidence, evidence)
+    results = [(skill, skill_confidences[skill], skill_evidence[skill]) for skill in skill_confidences]
+    results.sort(key=lambda x: x[1], reverse=True)
+    return results
 
-    return sorted(skill_confidences.items(), key=lambda x: x[1], reverse=True)
 
 # --- NEW FUNCTION ---
 def get_skill_embedding(skill_name):