Unstructured-IO · gyx09212214-prog · Jun 12, 2026 · Jun 19, 2026 · Jun 19, 2026 · Jun 19, 2026
diff --git a/test_unstructured/partition/html/test_partition.py b/test_unstructured/partition/html/test_partition.py
@@ -203,6 +203,43 @@ def test_partition_html_processes_chinese_chracters():
     assert elements[0].text == "每日新闻"
 
 
+def test_partition_html_preserves_numeric_table_cell_text_in_chunked_html():
+    html_text = """
+    <html>
+      <body>
+        <h1>Financial metrics</h1>
+        <table>
+          <tr><th>metric</th><th>value</th></tr>
+          <tr><td>revenue</td><td>478923</td></tr>
+          <tr><td>shares</td><td>1234567</td></tr>
+          <tr><td>ticker</td><td>000001</td></tr>
+          <tr><td>ratio</td><td>0.000123</td></tr>
+          <tr><td>market cap</td><td>999999999999999999</td></tr>
+        </table>
+      </body>
+    </html>
+    """
+
+    elements = partition_html(
+        text=html_text,
+        chunking_strategy="by_title",
+        max_characters=80,
+        new_after_n_chars=80,
+    )
+
+    table_html = "".join(
+        e.metadata.text_as_html or "" for e in elements if isinstance(e, TableChunk)
+    )
+
+    assert "<td>478923</td>" in table_html
+    assert "<td>1234567</td>" in table_html
+    assert "<td>000001</td>" in table_html
+    assert "<td>0.000123</td>" in table_html
+    assert "<td>999999999999999999</td>" in table_html
+    assert "e+" not in table_html.lower()
+    assert "e-" not in table_html.lower()
+
+
 def test_emoji_appears_with_emoji_utf8_code():
     assert partition_html(text='<html charset="utf-8"><p>Hello &#128512;</p></html>') == [
         Text("Hello 😀")

diff --git a/test_unstructured/partition/test_csv.py b/test_unstructured/partition/test_csv.py
@@ -114,6 +114,32 @@ def test_partition_csv_from_file_with_metadata_filename():
     assert elements[0].metadata.filename == "test"
 
 
+def test_partition_csv_preserves_numeric_cell_text_in_table_html(tmp_path):
+    file_path = tmp_path / "financial-metrics.csv"
+    file_path.write_text(
+        "metric,value\n"
+        "revenue,478923\n"
+        "shares,1234567\n"
+        "ticker,000001\n"
+        "ratio,0.000123\n"
+        "market cap,999999999999999999\n",
+        encoding="utf-8",
+    )
+
+    table = partition_csv(filename=str(file_path))[0]
+
+    assert table.metadata.text_as_html == (
+        "<table>"
+        "<tr><td>metric</td><td>value</td></tr>"
+        "<tr><td>revenue</td><td>478923</td></tr>"
+        "<tr><td>shares</td><td>1234567</td></tr>"
+        "<tr><td>ticker</td><td>000001</td></tr>"
+        "<tr><td>ratio</td><td>0.000123</td></tr>"
+        "<tr><td>market cap</td><td>999999999999999999</td></tr>"
+        "</table>"
+    )
+
+
 # -- .metadata.last_modified ---------------------------------------------------------------------
 
 

diff --git a/test_unstructured/partition/test_tsv.py b/test_unstructured/partition/test_tsv.py
@@ -73,6 +73,32 @@ def test_partition_tsv_from_file_with_metadata_filename():
     assert all(element.metadata.filename == "test" for element in elements)
 
 
+def test_partition_tsv_preserves_numeric_cell_text_in_table_html(tmp_path):
+    file_path = tmp_path / "financial-metrics.tsv"
+    file_path.write_text(
+        "metric\tvalue\n"
+        "revenue\t478923\n"
+        "shares\t1234567\n"
+        "ticker\t000001\n"
+        "ratio\t0.000123\n"
+        "market cap\t999999999999999999\n",
+        encoding="utf-8",
+    )
+
+    table = partition_tsv(filename=str(file_path), include_header=False)[0]
+
+    assert table.metadata.text_as_html == (
+        "<table>"
+        "<tr><td>metric</td><td>value</td></tr>"
+        "<tr><td>revenue</td><td>478923</td></tr>"
+        "<tr><td>shares</td><td>1234567</td></tr>"
+        "<tr><td>ticker</td><td>000001</td></tr>"
+        "<tr><td>ratio</td><td>0.000123</td></tr>"
+        "<tr><td>market cap</td><td>999999999999999999</td></tr>"
+        "</table>"
+    )
+
+
 # -- .metadata.last_modified ---------------------------------------------------------------------
 
 

diff --git a/unstructured/partition/csv.py b/unstructured/partition/csv.py
@@ -8,7 +8,7 @@
 import pandas as pd
 
 from unstructured.chunking import add_chunking_strategy
-from unstructured.common.html_table import HtmlTable
+from unstructured.common.html_table import HtmlTable, htmlify_matrix_of_cell_texts
 from unstructured.documents.elements import Element, ElementMetadata, Table
 from unstructured.file_utils.model import FileType
 from unstructured.partition.common.metadata import apply_metadata, get_last_modified_date
@@ -58,15 +58,19 @@ def partition_csv(
 
     csv.field_size_limit(CSV_FIELD_LIMIT)
     with ctx.open() as file:
-        read_kw: dict = {"header": ctx.header, "sep": ctx.delimiter, "encoding": ctx.encoding}
+        read_kw: dict = {
+            "header": ctx.header,
+            "sep": ctx.delimiter,
+            "encoding": ctx.encoding,
+            "dtype": str,
+            "keep_default_na": False,
+        }
         # sep=None is not supported by the C engine; use Python engine to avoid ParserWarning.
         if ctx.delimiter is None:
             read_kw["engine"] = "python"
         dataframe = pd.read_csv(file, **read_kw)
 
-    html_table = HtmlTable.from_html_text(
-        dataframe.to_html(index=False, header=include_header, na_rep="")
-    )
+    html_table = HtmlTable.from_html_text(_dataframe_to_html_text(dataframe, include_header))
 
     metadata = ElementMetadata(
         filename=filename,
@@ -78,6 +82,13 @@ def partition_csv(
     return [Table(text=html_table.text, metadata=metadata, detection_origin=DETECTION_ORIGIN)]
 
 
+def _dataframe_to_html_text(dataframe: pd.DataFrame, include_header: bool) -> str:
+    """Render a dataframe as table HTML without pandas numeric formatting."""
+    rows = dataframe.astype(str).values.tolist()
+    matrix = [[str(column) for column in dataframe.columns]] + rows if include_header else rows
+    return htmlify_matrix_of_cell_texts(matrix)
+
+
 class _CsvPartitioningContext:
     """Encapsulates the partitioning-run details.
 

diff --git a/unstructured/partition/tsv.py b/unstructured/partition/tsv.py
@@ -5,7 +5,7 @@
 import pandas as pd
 
 from unstructured.chunking import add_chunking_strategy
-from unstructured.common.html_table import HtmlTable
+from unstructured.common.html_table import HtmlTable, htmlify_matrix_of_cell_texts
 from unstructured.documents.elements import Element, ElementMetadata, Table
 from unstructured.file_utils.model import FileType
 from unstructured.partition.common.common import (
@@ -41,18 +41,22 @@ def partition_tsv(
 
     header = 0 if include_header else None
 
+    read_kw: dict[str, Any] = {
+        "sep": "\t",
+        "header": header,
+        "dtype": str,
+        "keep_default_na": False,
+    }
     if filename:
-        dataframe = pd.read_csv(filename, sep="\t", header=header)
+        dataframe = pd.read_csv(filename, **read_kw)
     else:
         assert file is not None
         # -- Note(scanny): `SpooledTemporaryFile` on Python<3.11 does not implement `.readable()`
         # -- which triggers an exception on `pd.DataFrame.read_csv()` call.
         f = spooled_to_bytes_io_if_needed(file)
-        dataframe = pd.read_csv(f, sep="\t", header=header)
+        dataframe = pd.read_csv(f, **read_kw)
 
-    html_table = HtmlTable.from_html_text(
-        dataframe.to_html(index=False, header=include_header, na_rep="")
-    )
+    html_table = HtmlTable.from_html_text(_dataframe_to_html_text(dataframe, include_header))
 
     metadata = ElementMetadata(
         filename=filename,
@@ -62,3 +66,10 @@ def partition_tsv(
     metadata.detection_origin = DETECTION_ORIGIN
 
     return [Table(text=html_table.text, metadata=metadata)]
+
+
+def _dataframe_to_html_text(dataframe: pd.DataFrame, include_header: bool) -> str:
+    """Render a dataframe as table HTML without pandas numeric formatting."""
+    rows = dataframe.astype(str).values.tolist()
+    matrix = [[str(column) for column in dataframe.columns]] + rows if include_header else rows
+    return htmlify_matrix_of_cell_texts(matrix)