feat: implement similarity engine for hybrid semantic and feature-based project comparison

src/similarity_model/similarity_engine.py

@@ -167,7 +167,21 @@ def compare_candidate(
         unique_query_count=unique_query_count
     )
 
-    hybrid_score = base_similarity
+    # Calculate Jaccard word overlap on description + abstract to detect direct copy-pastes
+    query_desc = (str(query_project.get("abstract", "")) + " " + str(query_project.get("description", ""))).strip()
+    candidate_desc = (str(candidate_row.get("abstract", "")) + " " + str(candidate_row.get("description", ""))).strip()
+    
+    words_q = set(normalize_text(query_desc).split())
+    words_c = set(normalize_text(candidate_desc).split())
+    
+    jaccard_overlap = 0.0
+    if words_q and words_c:
+        jaccard_overlap = len(words_q.intersection(words_c)) / len(words_q.union(words_c))
+
+    if jaccard_overlap >= 0.60:
+        hybrid_score = 0.95
+    else:
+        hybrid_score = base_similarity
 
     originality_score = compute_originality(
         hybrid_score=hybrid_score,