Update app.py

app.py

@@ -38,22 +38,29 @@ def extract_text_from_pdf(pdf_path):
         text_chunks.extend(chunks)
     return text_chunks
 
+# Kontexte nach Relevanz bewerten
+def rank_contexts_by_relevance(query, contexts):
+    scores = model.encode([query]) @ model.encode(contexts).T
+    ranked_contexts = sorted(zip(scores[0], contexts), key=lambda x: x[0], reverse=True)
+    return [context for _, context in ranked_contexts[:5]]  # Nur die Top 5 Kontexte
+
 # Suche und Bewertung
 def search_and_rank_answers(query, index, documents, k=10):
     query_embedding = model.encode([query])[0].astype('float32')
     D, I = index.search(np.array([query_embedding]), k=k)
 
+    ranked_contexts = [documents[i] for i in I[0]]
+    top_contexts = rank_contexts_by_relevance(query, ranked_contexts)
+
     ranked_answers = []
-    for i, doc_index in enumerate(I[0]):
-        context = documents[doc_index]
+    for context in top_contexts:
         try:
             result = qa_model(question=query, context=context)
-            ranked_answers.append((result['answer'], D[0][i]))  # (Antwort, Distanz)
+            ranked_answers.append((result['answer'], result['score']))
         except Exception as e:
             logging.warning(f"Fehler bei der Antwortgenerierung: {e}")
 
-    # Antworten nach ihrer Relevanz sortieren
-    ranked_answers = sorted(ranked_answers, key=lambda x: x[1])
+    ranked_answers = sorted(ranked_answers, key=lambda x: x[1], reverse=True)
     return [answer for answer, _ in ranked_answers]
 
 # Antworten kombinieren
@@ -76,6 +83,7 @@ def chatbot_response(pdf_path, question):
     # Antworten kombinieren
     detailed_answer = combine_answers(answers)
 
+    logging.info(f"Antwort: {detailed_answer}")
     return detailed_answer
 
 # Gradio-Interface