diff --git a/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/HanlpDictMatchStrategy.java b/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/HanlpDictMatchStrategy.java
index 139c23351..29a29b9b8 100644
--- a/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/HanlpDictMatchStrategy.java
+++ b/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/HanlpDictMatchStrategy.java
@@ -9,6 +9,7 @@ import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.stereotype.Service;
 
 import java.util.ArrayList;
+import java.util.Comparator;
 import java.util.LinkedHashSet;
 import java.util.List;
 import java.util.Set;
@@ -50,8 +51,11 @@ public class HanlpDictMatchStrategy extends SingleMatchStrategy<HanlpMapResult>
             return new ArrayList<>();
         }
         // step3. merge pre/suffix result
+        // sort by similarity (desc) first, then name length (desc), so that
+        // higher-similarity records are inserted first and survive LinkedHashSet dedup
         hanlpMapResults = hanlpMapResults.stream()
-                .sorted((a, b) -> -(b.getName().length() - a.getName().length()))
+                .sorted(Comparator.comparingDouble(HanlpMapResult::getSimilarity).reversed()
+                        .thenComparing((a, b) -> Integer.compare(b.getName().length(), a.getName().length())))
                 .collect(Collectors.toCollection(LinkedHashSet::new));
 
         // step4. filter by similarity
diff --git a/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/MapFilter.java b/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/MapFilter.java
index aad8d10c4..5851fccfd 100644
--- a/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/MapFilter.java
+++ b/headless/chat/src/main/java/com/tencent/supersonic/headless/chat/mapper/MapFilter.java
@@ -123,15 +123,9 @@ public class MapFilter {
                     .filter(SchemaElementMatch::isFullMatched).collect(Collectors.toList());
 
             if (!fullMatches.isEmpty()) {
-                // If there are objects with similarity=1.0, choose the one with the longest
-                // detectWord and smallest offset
-                SchemaElementMatch bestMatch = fullMatches.stream()
-                        .max(Comparator.comparing(
-                                (SchemaElementMatch match) -> match.getDetectWord().length()))
-                        .orElse(null);
-                if (bestMatch != null) {
-                    result.add(bestMatch);
-                }
+                // Keep all records with similarity=1.0, as they may correspond to different
+                // elementIds with the same detectWord
+                result.addAll(fullMatches);
             } else {
                 // If there are no objects with similarity=1.0, keep all objects with similarity<1.0
                 result.addAll(group);