diff --git a/fluid/PaddleRec/word2vec/train.py b/fluid/PaddleRec/word2vec/train.py
index 557411169e8bfafecc2fa701f639ba6e93cacfb3..6bac5e6af2ea009811ec7b991c69a2609fbd1541 100644
--- a/fluid/PaddleRec/word2vec/train.py
+++ b/fluid/PaddleRec/word2vec/train.py
@@ -78,7 +78,8 @@ def parse_args():
         help='max code length used by hierarchical sigmoid, (default: 40)')
     parser.add_argument(
         '--is_sparse',
-        type=bool,
+        action='store_true',
+        required=False,
         default=False,
         help='embedding and nce will use sparse or not, (default: False)')