diff --git a/paddlehub/datasets/base_nlp_dataset.py b/paddlehub/datasets/base_nlp_dataset.py
index c4cebdda83cd96595c41c480d700944292fb4061..c9e425ad7c3ccd212f90ccdf526e359f0f9d676a 100644
--- a/paddlehub/datasets/base_nlp_dataset.py
+++ b/paddlehub/datasets/base_nlp_dataset.py
@@ -276,7 +276,7 @@ class TextClassificationDataset(BaseNLPDataset, paddle.io.Dataset):
             if Version(paddlenlp.__version__) >= Version('2.0.0rc5'):
                 token_type_ids = np.array(record['token_type_ids'])
             else:
-                token_type_ids = record['segment_ids']
+                token_type_ids = np.array(record['segment_ids'])
 
             if 'label' in record.keys():
                 return input_ids, token_type_ids, np.array(record['label'], dtype=np.int64)