diff --git a/src/memos/mem_reader/read_multi_modal/base.py b/src/memos/mem_reader/read_multi_modal/base.py
index 95d427864..737a3fe1e 100644
--- a/src/memos/mem_reader/read_multi_modal/base.py
+++ b/src/memos/mem_reader/read_multi_modal/base.py
@@ -15,6 +15,7 @@
     TextualMemoryItem,
     TreeNodeTextualMemoryMetadata,
 )
+from memos.memories.textual.tree_text_memory.retrieve.retrieve_utils import FastTokenizer
 from memos.utils import timed
 
 from .utils import detect_lang, get_text_splitter
@@ -90,6 +91,7 @@ def __init__(self, embedder, llm=None):
         """
         self.embedder = embedder
         self.llm = llm
+        self.tokenizer = FastTokenizer(use_jieba=True, use_stopwords=True)
 
     @abstractmethod
     def create_source(
diff --git a/src/memos/mem_reader/read_multi_modal/file_content_parser.py b/src/memos/mem_reader/read_multi_modal/file_content_parser.py
index 00da08b1c..9f4ab94c2 100644
--- a/src/memos/mem_reader/read_multi_modal/file_content_parser.py
+++ b/src/memos/mem_reader/read_multi_modal/file_content_parser.py
@@ -710,7 +710,7 @@ def _make_fallback(
             chunk_idx: int, chunk_text: str, reason: str = "raw"
         ) -> TextualMemoryItem:
             """Create fallback memory item with raw chunk text."""
-            return _make_memory_item(
+            raw_chunk_mem = _make_memory_item(
                 value=chunk_text,
                 tags=[
                     "mode:fine",
@@ -721,6 +721,11 @@ def _make_fallback(
                 chunk_idx=chunk_idx,
                 chunk_content=chunk_text,
             )
+            tags_list = self.tokenizer.tokenize_mixed(raw_chunk_mem.metadata.key)
+            tags_list = [tag for tag in tags_list if len(tag) > 1]
+            tags_list = sorted(tags_list, key=len, reverse=True)
+            raw_chunk_mem.metadata.tags.extend(tags_list[:5])
+            return raw_chunk_mem
 
         # Handle empty chunks case
         if not valid_chunks: