New feature

Allow document backends to populate a generic metadata fields in the ConversionResult object.

For example: PDF metadata, USPTO patents metadata, etc

Specs

• ConversionResult:

  • Add metadata: Dict[str, Any] = {}

  • docling/docling/datamodel/document.py

    Lines 198 to 210 in ff351fd

    	class ConversionResult(BaseModel):
    	input: InputDocument
    	status: ConversionStatus = ConversionStatus.PENDING # failure, success
    	errors: List[ErrorItem] = [] # structure to keep errors
    	pages: List[Page] = []
    	assembled: AssembledUnit = AssembledUnit()
    	timings: Dict[str, ProfilingItem] = {}
    	confidence: ConfidenceReport = Field(default_factory=ConfidenceReport)
    	document: DoclingDocument = _EMPTY_DOCLING_DOC

• Backends

  • Add extract_metadata() to the AbstractDocumentBackend. It can just be a default implementation returning the empty dict.

  • docling/docling/backend/abstract_backend.py

    Lines 13 to 39 in ff351fd

    	class AbstractDocumentBackend(ABC):
    	@abstractmethod
    	def __init__(self, in_doc: "InputDocument", path_or_stream: Union[BytesIO, Path]):
    	self.file = in_doc.file
    	self.path_or_stream = path_or_stream
    	self.document_hash = in_doc.document_hash
    	self.input_format = in_doc.format
    	@abstractmethod
    	def is_valid(self) -> bool:
    	pass
    	@classmethod
    	@abstractmethod
    	def supports_pagination(cls) -> bool:
    	pass
    	def unload(self):
    	if isinstance(self.path_or_stream, BytesIO):
    	self.path_or_stream.close()
    	self.path_or_stream = None
    	@classmethod
    	@abstractmethod
    	def supported_formats(cls) -> Set["InputFormat"]:
    	pass

• Pipeline

  • Start with the SimplePipeline, add something like conv_res.document = conv_res.input._backend.extract_metadata()

  • docling/docling/pipeline/simple_pipeline.py

    Lines 39 to 40 in ff351fd

    	with TimeRecorder(conv_res, "doc_build", scope=ProfilingScope.DOCUMENT):
    	conv_res.document = conv_res.input._backend.convert()