diff --git a/src/graph/optimizer/rule/PushEFilterDownRule.cpp b/src/graph/optimizer/rule/PushEFilterDownRule.cpp
index 3b5df8904d2..b49ccfb5d1d 100644
--- a/src/graph/optimizer/rule/PushEFilterDownRule.cpp
+++ b/src/graph/optimizer/rule/PushEFilterDownRule.cpp
@@ -144,7 +144,7 @@ std::string PushEFilterDownRule::toString() const {
                                                             const storage::cpp2::EdgeProp &edge,
                                                             meta::SchemaManager *schemaMng,
                                                             ObjectPool *pool) {
-  auto edgeNameResult = schemaMng->toEdgeName(spaceId, edge.get_type());
+  auto edgeNameResult = schemaMng->toEdgeName(spaceId, std::abs(edge.get_type()));
   if (!edgeNameResult.ok()) {
     return nullptr;
   }
diff --git a/tests/tck/features/optimizer/PushEFilterDownRule.feature b/tests/tck/features/optimizer/PushEFilterDownRule.feature
index 96637782021..747e95fa6b8 100644
--- a/tests/tck/features/optimizer/PushEFilterDownRule.feature
+++ b/tests/tck/features/optimizer/PushEFilterDownRule.feature
@@ -25,6 +25,34 @@ Feature: Push EFilter down rule
       | 8  | Traverse  | 7            | {"edge filter": "", "filter": "(like.likeness==95)"} |
       | 7  | IndexScan | 0            |                                                      |
       | 0  | Start     |              |                                                      |
+    When profiling query:
+      """
+      MATCH (v:player{name: 'Tim Duncan'})<-[e:like{likeness: 95}]-() return v.player.name AS name
+      """
+    Then the result should be, in any order:
+      | name         |
+      | "Tim Duncan" |
+      | "Tim Duncan" |
+    And the execution plan should be:
+      | id | name      | dependencies | operator info                                        |
+      | 5  | Project   | 8            |                                                      |
+      | 8  | Traverse  | 7            | {"edge filter": "", "filter": "(like.likeness==95)"} |
+      | 7  | IndexScan | 0            |                                                      |
+      | 0  | Start     |              |                                                      |
+    When profiling query:
+      """
+      MATCH (v:player{name: 'Tim Duncan'})-[e:like{likeness: 95}]-() return v.player.name AS name
+      """
+    Then the result should be, in any order:
+      | name         |
+      | "Tim Duncan" |
+      | "Tim Duncan" |
+    And the execution plan should be:
+      | id | name      | dependencies | operator info                                        |
+      | 5  | Project   | 8            |                                                      |
+      | 8  | Traverse  | 7            | {"edge filter": "", "filter": "(like.likeness==95)"} |
+      | 7  | IndexScan | 0            |                                                      |
+      | 0  | Start     |              |                                                      |
     When profiling query:
       """
       MATCH (v:player{name: 'Tim Duncan'})-[e:like*1..2{likeness: 95}]->() return v.player.name AS name