deepdoc-lib 0.2.0__py3-none-any.whl

deepdoc/parser/docx_parser.py

@@ -0,0 +1,150 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+from docx import Document
+import re
+import pandas as pd
+from collections import Counter
+from io import BytesIO
+
+from ..config import TokenizerConfig
+from ..depend.rag_tokenizer import RagTokenizer
+
+
+class RAGFlowDocxParser:
+    def __init__(self, tokenizer_cfg: TokenizerConfig | None = None):
+        if tokenizer_cfg is None:
+            tokenizer_cfg = TokenizerConfig.from_env()
+        self.tokenizer_cfg = tokenizer_cfg
+        self.tokenizer = RagTokenizer(
+            dict_prefix=tokenizer_cfg.resolve_dict_prefix(),
+            offline=tokenizer_cfg.offline,
+            nltk_data_dir=tokenizer_cfg.nltk_data_dir,
+        )
+
+    def __extract_table_content(self, tb):
+        df = []
+        for row in tb.rows:
+            df.append([c.text for c in row.cells])
+        return self.__compose_table_content(pd.DataFrame(df))
+
+    def __compose_table_content(self, df):
+
+        def blockType(b):
+            pattern = [
+                ("^(20|19)[0-9]{2}[年/-][0-9]{1,2}[月/-][0-9]{1,2}日*$", "Dt"),
+                (r"^(20|19)[0-9]{2}年$", "Dt"),
+                (r"^(20|19)[0-9]{2}[年/-][0-9]{1,2}月*$", "Dt"),
+                ("^[0-9]{1,2}[月/-][0-9]{1,2}日*$", "Dt"),
+                (r"^第*[一二三四1-4]季度$", "Dt"),
+                (r"^(20|19)[0-9]{2}年*[一二三四1-4]季度$", "Dt"),
+                (r"^(20|19)[0-9]{2}[ABCDE]$", "DT"),
+                ("^[0-9.,+%/ -]+$", "Nu"),
+                (r"^[0-9A-Z/\._~-]+$", "Ca"),
+                (r"^[A-Z]*[a-z' -]+$", "En"),
+                (r"^[0-9.,+-]+[0-9A-Za-z/$￥%<>（）()' -]+$", "NE"),
+                (r"^.{1}$", "Sg")
+            ]
+            for p, n in pattern:
+                if re.search(p, b):
+                    return n
+            tks = [t for t in self.tokenizer.tokenize(b).split() if len(t) > 1]
+            if len(tks) > 3:
+                if len(tks) < 12:
+                    return "Tx"
+                else:
+                    return "Lx"
+
+            if len(tks) == 1 and self.tokenizer.tag(tks[0]) == "nr":
+                return "Nr"
+
+            return "Ot"
+
+        if len(df) < 2:
+            return []
+        max_type = Counter([blockType(str(df.iloc[i, j])) for i in range(
+            1, len(df)) for j in range(len(df.iloc[i, :]))])
+        max_type = max(max_type.items(), key=lambda x: x[1])[0]
+
+        colnm = len(df.iloc[0, :])
+        hdrows = [0]  # header is not necessarily appear in the first line
+        if max_type == "Nu":
+            for r in range(1, len(df)):
+                tys = Counter([blockType(str(df.iloc[r, j]))
+                              for j in range(len(df.iloc[r, :]))])
+                tys = max(tys.items(), key=lambda x: x[1])[0]
+                if tys != max_type:
+                    hdrows.append(r)
+
+        lines = []
+        for i in range(1, len(df)):
+            if i in hdrows:
+                continue
+            hr = [r - i for r in hdrows]
+            hr = [r for r in hr if r < 0]
+            t = len(hr) - 1
+            while t > 0:
+                if hr[t] - hr[t - 1] > 1:
+                    hr = hr[t:]
+                    break
+                t -= 1
+            headers = []
+            for j in range(len(df.iloc[i, :])):
+                t = []
+                for h in hr:
+                    x = str(df.iloc[i + h, j]).strip()
+                    if x in t:
+                        continue
+                    t.append(x)
+                t = ",".join(t)
+                if t:
+                    t += ": "
+                headers.append(t)
+            cells = []
+            for j in range(len(df.iloc[i, :])):
+                if not str(df.iloc[i, j]):
+                    continue
+                cells.append(headers[j] + str(df.iloc[i, j]))
+            lines.append(";".join(cells))
+
+        if colnm > 3:
+            return lines
+        return ["\n".join(lines)]
+
+    def __call__(self, fnm, from_page=0, to_page=100000000):
+        self.doc = Document(fnm) if isinstance(
+            fnm, str) else Document(BytesIO(fnm))
+        pn = 0 # parsed page
+        secs = [] # parsed contents
+        for p in self.doc.paragraphs:
+            if pn > to_page:
+                break
+
+            runs_within_single_paragraph = [] # save runs within the range of pages
+            for run in p.runs:
+                if pn > to_page:
+                    break
+                if from_page <= pn < to_page and p.text.strip():
+                    runs_within_single_paragraph.append(run.text) # append run.text first
+
+                # wrap page break checker into a static method
+                if 'lastRenderedPageBreak' in run._element.xml:
+                    pn += 1
+
+            secs.append(("".join(runs_within_single_paragraph), p.style.name if hasattr(p.style, 'name') else '')) # then concat run.text as part of the paragraph
+
+        tbls = [self.__extract_table_content(tb) for tb in self.doc.tables]
+        return secs, tbls

deepdoc/parser/excel_parser.py

@@ -0,0 +1,270 @@
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+import logging
+import re
+import sys
+from io import BytesIO
+
+import pandas as pd
+from openpyxl import Workbook, load_workbook
+from PIL import Image
+
+from ..depend.find_codec import find_codec
+
+# copied from `/openpyxl/cell/cell.py`
+ILLEGAL_CHARACTERS_RE = re.compile(r"[\000-\010]|[\013-\014]|[\016-\037]")
+
+
+class RAGFlowExcelParser:
+    @staticmethod
+    def _load_excel_to_workbook(file_like_object):
+        if isinstance(file_like_object, bytes):
+            file_like_object = BytesIO(file_like_object)
+
+        # Read first 4 bytes to determine file type
+        file_like_object.seek(0)
+        file_head = file_like_object.read(4)
+        file_like_object.seek(0)
+
+        if not (file_head.startswith(b"PK\x03\x04") or file_head.startswith(b"\xd0\xcf\x11\xe0")):
+            logging.info("Not an Excel file, converting CSV to Excel Workbook")
+
+            try:
+                file_like_object.seek(0)
+                df = pd.read_csv(file_like_object, on_bad_lines='skip')
+                return RAGFlowExcelParser._dataframe_to_workbook(df)
+
+            except Exception as e_csv:
+                raise Exception(f"Failed to parse CSV and convert to Excel Workbook: {e_csv}")
+
+        try:
+            return load_workbook(file_like_object, data_only=True)
+        except Exception as e:
+            logging.info(f"openpyxl load error: {e}, try pandas instead")
+            try:
+                file_like_object.seek(0)
+                try:
+                    dfs = pd.read_excel(file_like_object, sheet_name=None)
+                    return RAGFlowExcelParser._dataframe_to_workbook(dfs)
+                except Exception as ex:
+                    logging.info(f"pandas with default engine load error: {ex}, try calamine instead")
+                    file_like_object.seek(0)
+                    df = pd.read_excel(file_like_object, engine="calamine")
+                    return RAGFlowExcelParser._dataframe_to_workbook(df)
+            except Exception as e_pandas:
+                raise Exception(f"pandas.read_excel error: {e_pandas}, original openpyxl error: {e}")
+
+    @staticmethod
+    def _clean_dataframe(df: pd.DataFrame):
+        def clean_string(s):
+            if isinstance(s, str):
+                return ILLEGAL_CHARACTERS_RE.sub(" ", s)
+            return s
+
+        return df.apply(lambda col: col.map(clean_string))
+
+    @staticmethod
+    def _dataframe_to_workbook(df):
+        # if contains multiple sheets use _dataframes_to_workbook
+        if isinstance(df, dict) and len(df) > 1:
+            return RAGFlowExcelParser._dataframes_to_workbook(df)
+
+        df = RAGFlowExcelParser._clean_dataframe(df)
+        wb = Workbook()
+        ws = wb.active
+        ws.title = "Data"
+
+        for col_num, column_name in enumerate(df.columns, 1):
+            ws.cell(row=1, column=col_num, value=column_name)
+
+        for row_num, row in enumerate(df.values, 2):
+            for col_num, value in enumerate(row, 1):
+                ws.cell(row=row_num, column=col_num, value=value)
+
+        return wb
+    
+    @staticmethod
+    def _dataframes_to_workbook(dfs: dict):
+        wb = Workbook()
+        default_sheet = wb.active
+        wb.remove(default_sheet)
+        
+        for sheet_name, df in dfs.items():
+            df = RAGFlowExcelParser._clean_dataframe(df)
+            ws = wb.create_sheet(title=sheet_name)
+            for col_num, column_name in enumerate(df.columns, 1):
+                ws.cell(row=1, column=col_num, value=column_name)
+            for row_num, row in enumerate(df.values, 2):
+                for col_num, value in enumerate(row, 1):
+                    ws.cell(row=row_num, column=col_num, value=value)
+        return wb
+
+    @staticmethod
+    def _extract_images_from_worksheet(ws, sheetname=None):
+        """
+        Extract images from a worksheet and enrich them with vision-based descriptions.
+
+        Returns: List[dict]
+        """
+        images = getattr(ws, "_images", [])
+        if not images:
+            return []
+
+        raw_items = []
+
+        for img in images:
+            try:
+                img_bytes = img._data()
+                pil_img = Image.open(BytesIO(img_bytes)).convert("RGB")
+
+                anchor = img.anchor
+                if hasattr(anchor, "_from") and hasattr(anchor, "_to"):
+                    r1, c1 = anchor._from.row + 1, anchor._from.col + 1
+                    r2, c2 = anchor._to.row + 1, anchor._to.col + 1
+                    if r1 == r2 and c1 == c2:
+                        span = "single_cell"
+                    else:
+                        span = "multi_cell"
+                else:
+                    r1, c1 = anchor._from.row + 1, anchor._from.col + 1
+                    r2, c2 = r1, c1
+                    span = "single_cell"
+
+                item = {
+                    "sheet": sheetname or ws.title,
+                    "image": pil_img,
+                    "image_description": "",
+                    "row_from": r1,
+                    "col_from": c1,
+                    "row_to": r2,
+                    "col_to": c2,
+                    "span_type": span,
+                }
+                raw_items.append(item)
+            except Exception:
+                continue
+        return raw_items
+
+    def html(self, fnm, chunk_rows=256):
+        from html import escape
+
+        file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
+        wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)
+        tb_chunks = []
+
+        def _fmt(v):
+            if v is None:
+                return ""
+            return str(v).strip()
+
+        for sheetname in wb.sheetnames:
+            ws = wb[sheetname]
+            try:
+                rows = list(ws.rows)
+            except Exception as e:
+                logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
+                continue
+
+            if not rows:
+                continue
+
+            tb_rows_0 = "<tr>"
+            for t in list(rows[0]):
+                tb_rows_0 += f"<th>{escape(_fmt(t.value))}</th>"
+            tb_rows_0 += "</tr>"
+
+            for chunk_i in range((len(rows) - 1) // chunk_rows + 1):
+                tb = ""
+                tb += f"<table><caption>{sheetname}</caption>"
+                tb += tb_rows_0
+                for r in list(rows[1 + chunk_i * chunk_rows : min(1 + (chunk_i + 1) * chunk_rows, len(rows))]):
+                    tb += "<tr>"
+                    for i, c in enumerate(r):
+                        if c.value is None:
+                            tb += "<td></td>"
+                        else:
+                            tb += f"<td>{escape(_fmt(c.value))}</td>"
+                    tb += "</tr>"
+                tb += "</table>\n"
+                tb_chunks.append(tb)
+
+        return tb_chunks
+
+    def markdown(self, fnm):
+        import pandas as pd
+
+        file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
+        try:
+            file_like_object.seek(0)
+            df = pd.read_excel(file_like_object)
+        except Exception as e:
+            logging.warning(f"Parse spreadsheet error: {e}, trying to interpret as CSV file")
+            file_like_object.seek(0)
+            df = pd.read_csv(file_like_object, on_bad_lines='skip')
+        df = df.replace(r"^\s*$", "", regex=True)
+        return df.to_markdown(index=False)
+
+    def __call__(self, fnm):
+        file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
+        wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)
+
+        res = []
+        for sheetname in wb.sheetnames:
+            ws = wb[sheetname]
+            try:
+                rows = list(ws.rows)
+            except Exception as e:
+                logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
+                continue
+            if not rows:
+                continue
+            ti = list(rows[0])
+            for r in list(rows[1:]):
+                fields = []
+                for i, c in enumerate(r):
+                    if not c.value:
+                        continue
+                    t = str(ti[i].value) if i < len(ti) else ""
+                    t += ("：" if t else "") + str(c.value)
+                    fields.append(t)
+                line = "; ".join(fields)
+                if sheetname.lower().find("sheet") < 0:
+                    line += " ——" + sheetname
+                res.append(line)
+        return res
+
+    @staticmethod
+    def row_number(fnm, binary):
+        if fnm.split(".")[-1].lower().find("xls") >= 0:
+            wb = RAGFlowExcelParser._load_excel_to_workbook(BytesIO(binary))
+            total = 0
+            
+            for sheetname in wb.sheetnames:
+               try:
+                   ws = wb[sheetname]
+                   total += len(list(ws.rows))
+               except Exception as e:
+                   logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
+                   continue
+            return total
+
+        if fnm.split(".")[-1].lower() in ["csv", "txt"]:
+            encoding = find_codec(binary)
+            txt = binary.decode(encoding, errors="ignore")
+            return len(txt.split("\n"))
+
+
+if __name__ == "__main__":
+    psr = RAGFlowExcelParser()
+    psr(sys.argv[1])

deepdoc/parser/figure_parser.py

@@ -0,0 +1,182 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+from concurrent.futures import ThreadPoolExecutor, as_completed
+
+from PIL import Image
+
+from ..llm_adapter import LLMType, LLMAdapter
+from ..llm_adapter.vision import vision_llm_chunk as picture_vision_llm_chunk
+from ..depend.prompts import vision_llm_figure_describe_prompt
+
+# Try to import timeout from common, fallback to local
+try:
+    from ..common.connection_utils import timeout
+except ImportError:
+    from ..depend.timeout import timeout
+
+
+def vision_figure_parser_figure_data_wrapper(figures_data_without_positions):
+    if not figures_data_without_positions:
+        return []
+    return [
+        (
+            (figure_data[1], [figure_data[0]]),
+            [(0, 0, 0, 0, 0)],
+        )
+        for figure_data in figures_data_without_positions
+        if isinstance(figure_data[1], Image.Image)
+    ]
+
+
+def vision_figure_parser_docx_wrapper(sections, tbls, callback=None,**kwargs):
+    if not sections:
+        return tbls
+    try:
+        vision_model = LLMAdapter(kwargs.get("tenant_id"), LLMType.IMAGE2TEXT)
+        callback(0.7, "Visual model detected. Attempting to enhance figure extraction...")
+    except Exception:
+        vision_model = None
+    if vision_model:
+        figures_data = vision_figure_parser_figure_data_wrapper(sections)
+        try:
+            docx_vision_parser = VisionFigureParser(vision_model=vision_model, figures_data=figures_data, **kwargs)
+            boosted_figures = docx_vision_parser(callback=callback)
+            tbls.extend(boosted_figures)
+        except Exception as e:
+            callback(0.8, f"Visual model error: {e}. Skipping figure parsing enhancement.")
+    return tbls
+
+def vision_figure_parser_figure_xlsx_wrapper(images,callback=None, **kwargs):
+    tbls = []
+    if not images:
+        return []
+    try:
+        vision_model = LLMAdapter(kwargs.get("tenant_id"), LLMType.IMAGE2TEXT)
+        callback(0.2, "Visual model detected. Attempting to enhance Excel image extraction...")
+    except Exception:
+        vision_model = None
+    if vision_model:
+        figures_data = [((
+                        img["image"],   # Image.Image
+                        [img["image_description"]]     # description list (must be list)
+                    ),
+                    [
+                        (0, 0, 0, 0, 0)   # dummy position
+                    ]) for img in images]
+        try:
+            parser = VisionFigureParser(vision_model=vision_model, figures_data=figures_data, **kwargs)
+            callback(0.22, "Parsing images...")
+            boosted_figures = parser(callback=callback)
+            tbls.extend(boosted_figures)
+        except Exception as e:
+            callback(0.25, f"Excel visual model error: {e}. Skipping vision enhancement.")
+    return tbls
+
+def vision_figure_parser_pdf_wrapper(tbls, callback=None, **kwargs):
+    if not tbls:
+        return []
+    try:
+        vision_model = LLMAdapter(kwargs.get("tenant_id"), LLMType.IMAGE2TEXT)
+        callback(0.7, "Visual model detected. Attempting to enhance figure extraction...")
+    except Exception:
+        vision_model = None
+    if vision_model:
+        def is_figure_item(item):
+            return (
+                isinstance(item[0][0], Image.Image) and
+                isinstance(item[0][1], list)
+            )
+        figures_data = [item for item in tbls if is_figure_item(item)]
+        try:
+            docx_vision_parser = VisionFigureParser(vision_model=vision_model, figures_data=figures_data, **kwargs)
+            boosted_figures = docx_vision_parser(callback=callback)
+            tbls = [item for item in tbls if not is_figure_item(item)]
+            tbls.extend(boosted_figures)
+        except Exception as e:
+            callback(0.8, f"Visual model error: {e}. Skipping figure parsing enhancement.")
+    return tbls
+
+
+shared_executor = ThreadPoolExecutor(max_workers=10)
+
+
+class VisionFigureParser:
+    def __init__(self, vision_model, figures_data, *args, **kwargs):
+        self.vision_model = vision_model
+        self._extract_figures_info(figures_data)
+        assert len(self.figures) == len(self.descriptions)
+        assert not self.positions or (len(self.figures) == len(self.positions))
+
+    def _extract_figures_info(self, figures_data):
+        self.figures = []
+        self.descriptions = []
+        self.positions = []
+
+        for item in figures_data:
+            # position
+            if len(item) == 2 and isinstance(item[0], tuple) and len(item[0]) == 2 and isinstance(item[1], list) and isinstance(item[1][0], tuple) and len(item[1][0]) == 5:
+                img_desc = item[0]
+                assert len(img_desc) == 2 and isinstance(img_desc[0], Image.Image) and isinstance(img_desc[1], list), "Should be (figure, [description])"
+                self.figures.append(img_desc[0])
+                self.descriptions.append(img_desc[1])
+                self.positions.append(item[1])
+            else:
+                assert len(item) == 2 and isinstance(item[0], Image.Image) and isinstance(item[1], list), f"Unexpected form of figure data: get {len(item)=}, {item=}"
+                self.figures.append(item[0])
+                self.descriptions.append(item[1])
+
+    def _assemble(self):
+        self.assembled = []
+        self.has_positions = len(self.positions) != 0
+        for i in range(len(self.figures)):
+            figure = self.figures[i]
+            desc = self.descriptions[i]
+            pos = self.positions[i] if self.has_positions else None
+
+            figure_desc = (figure, desc)
+
+            if pos is not None:
+                self.assembled.append((figure_desc, pos))
+            else:
+                self.assembled.append((figure_desc,))
+
+        return self.assembled
+
+    def __call__(self, **kwargs):
+        callback = kwargs.get("callback", lambda prog, msg: None)
+
+        @timeout(30, 3)
+        def process(figure_idx, figure_binary):
+            description_text = picture_vision_llm_chunk(
+                binary=figure_binary,
+                vision_model=self.vision_model,
+                prompt=vision_llm_figure_describe_prompt(),
+                callback=callback,
+            )
+            return figure_idx, description_text
+
+        futures = []
+        for idx, img_binary in enumerate(self.figures or []):
+            futures.append(shared_executor.submit(process, idx, img_binary))
+
+        for future in as_completed(futures):
+            figure_num, txt = future.result()
+            if txt:
+                self.descriptions[figure_num] = txt + "\n".join(self.descriptions[figure_num])
+
+        self._assemble()
+
+        return self.assembled