pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2026-02-01 16:15:07 +08:00
Code Issues Packages Projects Releases Wiki Activity

Feat: support context window for docx (#12455)

Browse Source 
### What problem does this PR solve?

Feat: support context window for docx

#12303

Done:
- [x] naive.py
- [x] one.py

TODO:
- [ ] book.py
- [ ] manual.py

Fix: incorrect image position
Fix: incorrect chunk type tag

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
This commit is contained in:
Magicbook1108
2026-01-07 15:08:17 +08:00
committed by GitHub
parent a442c9cac6
commit 011bbe9556
7 changed files with 397 additions and 120 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
rag/app/book.py
View File

@ -87,10 +87,18 @@ def chunk(filename, binary=None, from_page=0, to_page=100000,
callback(0.1, "Start to parse.")
doc_parser = naive.Docx()
# TODO: table of contents need to be removed
sections, tbls = doc_parser(
main_sections = doc_parser(
filename, binary=binary, from_page=from_page, to_page=to_page)
sections = []
tbls = []
for text, image, html in main_sections:
sections.append((text, image))
tbls.append(((None, html), ""))
remove_contents_table(sections, eng=is_english(
random_choices([t for t, _ in sections], k=200)))
tbls = vision_figure_parser_docx_wrapper(sections=sections, tbls=tbls, callback=callback, **kwargs)
# tbls = [((None, lns), None) for lns in tbls]
sections = [(item[0], item[1] if item[1] is not None else "") for item in sections if

191
rag/app/naive.py
View File

@ -23,6 +23,8 @@ from timeit import default_timer as timer
from docx import Document
from docx.image.exceptions import InvalidImageStreamError, UnexpectedEndOfFileError, UnrecognizedImageError
from docx.opc.pkgreader import _SerializedRelationships, _SerializedRelationship
from docx.table import Table as DocxTable
from docx.text.paragraph import Paragraph
from docx.opc.oxml import parse_xml
from markdown import markdown
from PIL import Image
@ -33,15 +35,15 @@ from api.db.services.llm_service import LLMBundle
from rag.utils.file_utils import extract_embed_file, extract_links_from_pdf, extract_links_from_docx, extract_html
from deepdoc.parser import DocxParser, ExcelParser, HtmlParser, JsonParser, MarkdownElementExtractor, MarkdownParser, \
PdfParser, TxtParser
from deepdoc.parser.figure_parser import VisionFigureParser, vision_figure_parser_docx_wrapper, \
from deepdoc.parser.figure_parser import VisionFigureParser, vision_figure_parser_docx_wrapper_naive, \
vision_figure_parser_pdf_wrapper
from deepdoc.parser.pdf_parser import PlainParser, VisionParser
from deepdoc.parser.docling_parser import DoclingParser
from deepdoc.parser.tcadp_parser import TCADPParser
from common.parser_config_utils import normalize_layout_recognizer
from rag.nlp import concat_img, find_codec, naive_merge, naive_merge_with_images, naive_merge_docx, rag_tokenizer, \
tokenize_chunks, tokenize_chunks_with_images, tokenize_table, attach_media_context, append_context2table_image4pdf
tokenize_chunks, doc_tokenize_chunks_with_images, tokenize_table, append_context2table_image4pdf, tokenize_chunks_with_images, \
attach_media_context # noqa: F401
def by_deepdoc(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", callback=None, pdf_cls=None,
**kwargs):
@ -343,67 +345,116 @@ class Docx(DocxParser):
pn = 0
lines = []
last_image = None
for p in self.doc.paragraphs:
table_idx = 0
def flush_last_image():
nonlocal last_image, lines
if last_image is not None:
lines.append({"text": "", "image": last_image, "table": None, "style": "Image"})
last_image = None
for block in self.doc._element.body:
if pn > to_page:
break
if from_page <= pn < to_page:
if p.text.strip():
if p.style and p.style.name == 'Caption':
former_image = None
if lines and lines[-1][1] and lines[-1][2] != 'Caption':
former_image = lines[-1][1].pop()
elif last_image:
former_image = last_image
last_image = None
lines.append((self.__clean(p.text), [former_image], p.style.name))
if block.tag.endswith('p'):
p = Paragraph(block, self.doc)
if from_page <= pn < to_page:
text = p.text.strip()
style_name = p.style.name if p.style else ""
if text:
if style_name == "Caption":
former_image = None
if lines and lines[-1].get("image") and lines[-1].get("style") != "Caption":
former_image = lines[-1].get("image")
lines.pop()
elif last_image is not None:
former_image = last_image
last_image = None
lines.append(
{
"text": self.__clean(text),
"image": former_image if former_image else None,
"table": None,
}
)
else:
flush_last_image()
lines.append(
{
"text": self.__clean(text),
"image": None,
"table": None,
}
)
current_image = self.get_picture(self.doc, p)
if current_image is not None:
lines.append(
{
"text": "",
"image": current_image,
"table": None,
}
)
else:
current_image = self.get_picture(self.doc, p)
image_list = [current_image]
if last_image:
image_list.insert(0, last_image)
last_image = None
lines.append((self.__clean(p.text), image_list, p.style.name if p.style else ""))
else:
if current_image := self.get_picture(self.doc, p):
if lines:
lines[-1][1].append(current_image)
else:
if current_image is not None:
last_image = current_image
for run in p.runs:
if 'lastRenderedPageBreak' in run._element.xml:
pn += 1
continue
if 'w:br' in run._element.xml and 'type="page"' in run._element.xml:
pn += 1
new_line = [(line[0], reduce(concat_img, line[1]) if line[1] else None) for line in lines]
tbls = []
for i, tb in enumerate(self.doc.tables):
title = self.__get_nearest_title(i, filename)
html = "<table>"
if title:
html += f"<caption>Table Location: {title}</caption>"
for r in tb.rows:
html += "<tr>"
i = 0
try:
while i < len(r.cells):
span = 1
c = r.cells[i]
for j in range(i + 1, len(r.cells)):
if c.text == r.cells[j].text:
span += 1
i = j
else:
break
i += 1
html += f"<td>{c.text}</td>" if span == 1 else f"<td colspan='{span}'>{c.text}</td>"
except Exception as e:
logging.warning(f"Error parsing table, ignore: {e}")
html += "</tr>"
html += "</table>"
tbls.append(((None, html), ""))
return new_line, tbls
for run in p.runs:
xml = run._element.xml
if "lastRenderedPageBreak" in xml:
pn += 1
continue
if "w:br" in xml and 'type="page"' in xml:
pn += 1
elif block.tag.endswith('tbl'):
if pn < from_page or pn > to_page:
table_idx += 1
continue
flush_last_image()
tb = DocxTable(block, self.doc)
title = self.__get_nearest_title(table_idx, filename)
html = "<table>"
if title:
html += f"<caption>Table Location: {title}</caption>"
for r in tb.rows:
html += "<tr>"
col_idx = 0
try:
while col_idx < len(r.cells):
span = 1
c = r.cells[col_idx]
for j in range(col_idx + 1, len(r.cells)):
if c.text == r.cells[j].text:
span += 1
col_idx = j
else:
break
col_idx += 1
html += f"<td>{c.text}</td>" if span == 1 else f"<td colspan='{span}'>{c.text}</td>"
except Exception as e:
logging.warning(f"Error parsing table, ignore: {e}")
html += "</tr>"
html += "</table>"
lines.append({"text": "", "image": None, "table": html})
table_idx += 1
flush_last_image()
new_line = [(line.get("text"), line.get("image"), line.get("table")) for line in lines]
return new_line
def to_markdown(self, filename=None, binary=None, inline_images: bool = True):
"""
@ -727,26 +778,26 @@ def chunk(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", ca
# fix "There is no item named 'word/NULL' in the archive", referring to https://github.com/python-openxml/python-docx/issues/1105#issuecomment-1298075246
_SerializedRelationships.load_from_xml = load_from_xml_v2
sections, tables = Docx()(filename, binary)
tables = vision_figure_parser_docx_wrapper(sections=sections, tbls=tables, callback=callback, **kwargs)
res = tokenize_table(tables, doc, is_english)
callback(0.8, "Finish parsing.")
st = timer()
# sections = (text, image, tables)
sections = Docx()(filename, binary)
# chunks list[dict]
# images list - index of image chunk in chunks
chunks, images = naive_merge_docx(
sections, int(parser_config.get(
"chunk_token_num", 128)), parser_config.get(
"delimiter", "\n!?。；！？"))
"delimiter", "\n!?。；！？"), table_context_size, image_context_size)
vision_figure_parser_docx_wrapper_naive(chunks=chunks, idx_lst=images, callback=callback, **kwargs)
res.extend(tokenize_chunks_with_images(chunks, doc, is_english, images, child_delimiters_pattern=child_deli))
callback(0.8, "Finish parsing.")
st = timer()
res.extend(doc_tokenize_chunks_with_images(chunks, doc, is_english, child_delimiters_pattern=child_deli))
logging.info("naive_merge({}): {}".format(filename, timer() - st))
res.extend(embed_res)
res.extend(url_res)
if table_context_size or image_context_size:
attach_media_context(res, table_context_size, image_context_size)
return res
elif re.search(r"\.pdf$", filename, re.IGNORECASE):
@ -1012,7 +1063,7 @@ def chunk(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", ca
res.extend(embed_res)
if url_res:
res.extend(url_res)
#if table_context_size or image_context_size:
# if table_context_size or image_context_size:
# attach_media_context(res, table_context_size, image_context_size)
return res

27
rag/app/one.py
View File

@ -22,7 +22,7 @@ from deepdoc.parser.utils import get_text
from rag.app import naive
from rag.nlp import rag_tokenizer, tokenize
from deepdoc.parser import PdfParser, ExcelParser, HtmlParser
from deepdoc.parser.figure_parser import vision_figure_parser_docx_wrapper
from deepdoc.parser.figure_parser import vision_figure_parser_docx_wrapper_naive
from rag.app.naive import by_plaintext, PARSERS
from common.parser_config_utils import normalize_layout_recognizer
@ -76,11 +76,26 @@ def chunk(filename, binary=None, from_page=0, to_page=100000,
if re.search(r"\.docx$", filename, re.IGNORECASE):
callback(0.1, "Start to parse.")
sections, tbls = naive.Docx()(filename, binary)
tbls = vision_figure_parser_docx_wrapper(sections=sections, tbls=tbls, callback=callback, **kwargs)
sections = [s for s, _ in sections if s]
for (_, html), _ in tbls:
sections.append(html)
sections = naive.Docx()(filename, binary)
cks = []
image_idxs = []
for text, image, table in sections:
if table is not None:
text = (text or "") + str(table)
ck_type = "table"
else:
ck_type = "image" if image is not None else "text"
if ck_type == "image":
image_idxs.append(len(cks))
cks.append({"text": text, "image": image, "ck_type": ck_type})
vision_figure_parser_docx_wrapper_naive(cks, image_idxs, callback, **kwargs)
for ck in cks:
print(ck)
sections = [ck["text"] for ck in cks if ck.get("text")]
callback(0.8, "Finish parsing.")
elif re.search(r"\.pdf$", filename, re.IGNORECASE):