githubmirror/UFO | Gitverse

learner/xml_loader.py
86 строк2 KB
Ray Huang
Merged PR 1532: cloned from https://github.com/microsoft/UFO/tree/vyokky/dev
23 май 2025, 05:11
23 май 2025, 05:11320a02b
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT License.

from . import basic
import os
from langchain_community.document_loaders import UnstructuredXMLLoader
from langchain.docstore.document import Document
import xml.etree.ElementTree as ET


class XMLLoader(basic.BasicDocumentLoader):
    """
    Class to load XML documents.
    """

    def __init__(self, directory: str = None):
        """
        Create a new XMLLoader.
        """

        super().__init__()
        self.extensions = ".xml"
        self.directory = directory

    def get_microsoft_document_metadata(self, file: str):
        """
        Get the metadata for the given file.
        :param file: The file to get the metadata for.
        :return: The metadata for the given file.
        """

        if not os.path.exists(file):
            return {"title": os.path.basename(file), "summary": os.path.basename(file)}

        tree = ET.parse(file)
        root = tree.getroot()

        # Extracting title
        if root.find("title") is not None:
            title = root.find("title").text
        else:
            title = None

        # Extracting content summary

        if root.find("Content-Summary") is not None:
            summary = root.find("Content-Summary").attrib["value"]
        else:
            summary = None

        return {"title": title, "summary": summary}

    def get_microsoft_document_text(self, file: str):
        """
        Get the text for the given file.
        :param file: The file to get the text for.
        :return: The text for the given file.
        """

        try:
            doc_text = UnstructuredXMLLoader(file).load()[0].page_content
        except:
            doc_text = None

        return doc_text

    def construct_document(self):
        """
        Construct a langchain document list.
        :return: The langchain document list.
        """
        documents = []
        for file in self.load_file_name():
            text = self.get_microsoft_document_text(file)
            metadata = self.get_microsoft_document_metadata(file + ".meta")
            title = metadata["title"]
            summary = metadata["summary"]
            page_content = """{title} - {summary}""".format(
                title=title, summary=summary
            )

            metadata = {"title": title, "summary": summary, "text": text}
            document = Document(page_content=page_content, metadata=metadata)

            documents.append(document)
        return documents