Upload app.py

app.py

@@ -759,11 +759,19 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
         
         top_k_retrieval = min(50, faiss_index.ntotal) if faiss_index else 50
         
+        # Criar mapeamento de scores por índice
+        retrieval_scores = {}
+        
         if faiss_index:
             # FAISS search - retorna (distances, indices)
-            scores, top_indices = faiss_index.search(query_normalized.astype('float32'), top_k_retrieval)
-            scores = scores[0]  # Flatten
+            faiss_scores, top_indices = faiss_index.search(query_normalized.astype('float32'), top_k_retrieval)
+            faiss_scores = faiss_scores[0]  # Flatten
             top_indices = top_indices[0]
+            scores = faiss_scores
+            # Mapear índices para scores
+            for i, idx in enumerate(top_indices):
+                if idx >= 0:
+                    retrieval_scores[int(idx)] = float(faiss_scores[i])
             logging.info(f"FAISS search: top score = {scores[0]:.3f}")
         else:
             # Fallback para busca bruta se não tiver FAISS
@@ -771,6 +779,9 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
             similarities = cosine_similarity(query_embedding, corpus_embeddings)[0]
             top_indices = np.argsort(similarities)[-top_k_retrieval:][::-1]
             scores = similarities[top_indices]
+            # Mapear índices para scores
+            for i, idx in enumerate(top_indices):
+                retrieval_scores[int(idx)] = float(similarities[idx])
         
         candidate_docs = []
         candidate_indices = []
@@ -807,7 +818,7 @@ async def search_api(query: str = Form(...), job_id: str = Form(...)):
             final_results.append({
                 "index": idx,
                 "score": float(score),
-                "cosine_score": float(similarities[idx]),
+                "cosine_score": retrieval_scores.get(idx, 0.0),
                 "citation_id": rank + 1
             })