Update functions.py

functions.py

@@ -109,33 +109,45 @@ def split_text_by_token_limit(text, tokenizer, max_tokens):
     return chunks
 
 
-def safe_sentiment(sentiment_pipeline, text):
+# def safe_sentiment(sentiment_pipeline, text):
+#     try:
+#         result = sentiment_pipeline(text)[0]
+#     except Exception as e:
+#         result = None
+#     return result
+
+
+def safe_sentiment_batch(sentiment_pipeline, texts):
+    """
+    breack donwn into batch
+    """
     try:
-        result = sentiment_pipeline(text)[0]
+        results = sentiment_pipeline(texts)
     except Exception as e:
-        result = None
-    return result
-
+        results = [None] * len(texts)
+    return results
 
 def analyze_detail(text, tokenizer, sentiment_pipeline, max_tokens):
     text = preprocess_text(text)
     chunks = split_text_by_token_limit(text, tokenizer, max_tokens)
     if not chunks:
         return None
-    # Initialize accumulated scores for each sentiment category
+    # ここでバッチ処理を実行（チャンク全体を一括推論）
+    results = safe_sentiment_batch(sentiment_pipeline, chunks)
+    
+    # 各チャンクの結果を集計
     scores = {"POSITIVE": 0, "NEGATIVE": 0, "NEUTRAL": 0}
-    for chunk in chunks:
-        result = safe_sentiment(sentiment_pipeline, chunk)
+    for result in results:
         if result is not None:
             label = result['label'].upper()
             if label in scores:
                 scores[label] += result['score']
     final_label = max(scores, key=lambda k: scores[k])
     final_score = scores[final_label]
+    
     return {"label": final_label, "score": final_score}
 
 
-
 def preprocess_text(text):
     # Replace URLs and user mentions
     text = re.sub(r'http\S+', 'http', text)