diff --git a/paddle/operators/rank_loss_op.cc b/paddle/operators/rank_loss_op.cc
index 14cddb609fb90d0d69da0fa8930ded16b3f1edb2..66571bd9a6378463d011d17f28532620e787588b 100644
--- a/paddle/operators/rank_loss_op.cc
+++ b/paddle/operators/rank_loss_op.cc
@@ -48,9 +48,9 @@ class RankLossOpMaker : public framework::OpProtoAndCheckerMaker {
   RankLossOpMaker(framework::OpProto *proto,
                   framework::OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
-    AddInput("P", "The first input of RankLoss operator.");
-    AddInput("Oi", "The second input of RankLoss operator.");
-    AddInput("Oj", "The third input of RankLoss operator.");
+    AddInput("P", "The desired target values for posteriors.");
+    AddInput("Oi", "The model output for item i.");
+    AddInput("Oj", "The model output for item j.");
     AddOutput("Out", "The output tensor of RankLoss operator.");
     AddComment(R"DOC(RankLoss operator
 
@@ -63,6 +63,8 @@ three inputs: P, Oi, and Oj, and the rank cost can be expressed as
   \tilde{P_{i,j}} = \left \{0, 0.5, 1 \right \} \ or \ \left \{0, 1 \right \}
 \f]
 
+A detailed explanation about these notations can be found in
+
 [1]. Chris Burges, Tal Shaked, Erin Renshaw, et al. Learning to
      Rank useing Gradient Descent.
 )DOC");