Portable Document Format (PDF), a file format standardized by ISO 32000, was developed by Adobe in 1992 for presenting documents, which include text formatting and images in a way that is independent of application software, hardware, and operating systems.
This guide covers how to load a PDF document into the LangChain Document format. This format will be used downstream.
LangChain integrates with a variety of PDF parsers. Some are simple and relatively low-level, while others support OCR and image processing or perform advanced document layout analysis.
The right choice depends on your application.
We will demonstrate these approaches on a sample file. Download the sample file and copy it to your data folder.
FILE_PATH ="./data/layout-parser-paper.pdf"
defshow_metadata(docs):if docs:print("[metadata]")print(list(docs[0].metadata.keys()))print("\n[examples]") max_key_length =max(len(k) for k in docs[0].metadata.keys())for k, v in docs[0].metadata.items():print(f"{k:<{max_key_length}} : {v}")
PyPDF
PyPDF is one of the most widely used Python libraries for PDF processing.
Here we use pypdf to load the PDF as an array of documents, each with a page number and containing page content and metadata.
LangChain's PyPDFLoader integrates with PyPDF to parse PDF documents into LangChain Document objects.
from langchain_community.document_loaders import PyPDFLoader# Initialize the PDF loaderloader =PyPDFLoader(FILE_PATH)# Load data into Document objectsdocs = loader.load()# Print the contents of the documentprint(docs[10].page_content[:300])
LayoutParser: A Unified Toolkit for DL-Based DIA 11
focuses on precision, efficiency, and robustness. The target documents may have
complicated structures, and may require training multiple layout detection models
to achieve the optimal accuracy. Light-weight pipelines are built for relatively
simple d
The load_and_split() method allows customizing how documents are chunked by passing a text splitter object, making it more flexible for different use cases.
from langchain_text_splitters import RecursiveCharacterTextSplitter# Load Documents and split into chunks. Chunks are returned as Documents.text_splitter =RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=200)docs = loader.load_and_split(text_splitter=text_splitter)print(docs[0].page_content)
LayoutParser: A Unified Toolkit for Deep
Learning Based Document Image Analysis
Zejiang Shen1 (� ), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain
Lee4, Jacob Carlson3, and Weining Li5
PyPDF(OCR)
Some PDFs contain text images within scanned documents or pictures. You can also use the rapidocr-onnxruntime package to extract text from images.
# Initialize PDF loader, enable image extraction optionloader =PyPDFLoader(FILE_PATH, extract_images=True)# load PDF pagedocs = loader.load()# access page contentprint(docs[4].page_content[:300])
LayoutParser: A Unified Toolkit for DL-Based DIA 5
Table 1: Current layout detection models in the LayoutParser model zoo
Dataset Base Model1 Large ModelNotes
PubLayNet [38] F / M M Layouts of modern scientific documents
PRImA [3] M - Layouts of scanned modern magazines and scientific reports
Newspaper
from langchain_community.document_loaders import PyPDFDirectoryLoader# directory pathloader =PyPDFDirectoryLoader("./data/")# load documentsdocs = loader.load()# print the number of documentsdocs_len =len(docs)print(docs_len)# get document from a directorydocument = docs[docs_len -1]
16
# print the contents of the documentprint(document.page_content[:300])
16 Z. Shen et al.
[23] Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z.,
Desmaison, A., Antiga, L., Lerer, A.: Automatic differentiation in pytorch (2017)
[24] Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen,
T., Lin, Z., Gimelshein, N., An
PyMuPDF is speed optimized and includes detailed metadata about the PDF and its pages. It returns one document per page.
LangChain's PyMuPDFLoader integrates with PyMuPDF to parse PDF documents into LangChain Document objects.
from langchain_community.document_loaders import PyMuPDFLoader# create an instance of the PyMuPDF loaderloader =PyMuPDFLoader(FILE_PATH)# load the documentdocs = loader.load()# print the contents of the documentprint(docs[10].page_content[:300])
LayoutParser: A Unified Toolkit for DL-Based DIA
11
focuses on precision, efficiency, and robustness. The target documents may have
complicated structures, and may require training multiple layout detection models
to achieve the optimal accuracy. Light-weight pipelines are built for relatively
simple d
Unstructured is a powerful library designed to handle various unstructured and semi-structured document formats. It excels at automatically identifying and categorizing different components within documents. Currently supports loading text files, PowerPoints, HTML, PDFs, images, and more.
LangChain's UnstructuredPDFLoader integrates with Unstructured to parse PDF documents into LangChain Document objects.
from langchain_community.document_loaders import UnstructuredPDFLoader# create an instance of UnstructuredPDFLoaderloader =UnstructuredPDFLoader(FILE_PATH)# load the datadocs = loader.load()# print the contents of the documentprint(docs[0].page_content[:300])
Matplotlib is building the font cache; this may take a moment.
1 2 0 2
n u J
1 2
]
V C . s c [
2 v 8 4 3 5 1 . 3 0 1 2 : v i X r a
LayoutParser: A Unified Toolkit for Deep Learning Based Document Image Analysis
Zejiang Shen1 ((cid:0)), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain Lee4, Jacob Carlson3, and Weining Li5
1 Allen Institute for AI s
Internally, unstructured creates different "elements" for each chunk of text. By default, these are combined, but can be easily separated by specifying mode="elements".
# Create an instance of UnstructuredPDFLoader (mode="elements”)loader =UnstructuredPDFLoader(FILE_PATH, mode="elements")# load the datadocs = loader.load()# print the contents of the documentprint(docs[0].page_content)
1 2 0 2
See the full set of element types for this particular article.
set(doc.metadata["category"] for doc in docs)# extract data categories
LangChain's PyPDFium2Loader integrates with PyPDFium2 to parse PDF documents into LangChain Document objects.
from langchain_community.document_loaders import PyPDFium2Loader# create an instance of the PyPDFium2 loaderloader =PyPDFium2Loader(FILE_PATH)# load datadocs = loader.load()# print the contents of the documentprint(docs[10].page_content[:300])
LayoutParser: A Unified Toolkit for DL-Based DIA 11
focuses on precision, efficiency, and robustness. The target documents may have
complicated structures, and may require training multiple layout detection models
to achieve the optimal accuracy. Light-weight pipelines are built for relatively
s
Note: When using PyPDFium2Loader, you may notice warning messages related to get_text_range(). These warnings are part of the library's internal operations and do not affect the PDF processing functionality. You can safely proceed with the tutorial despite these warnings, as they are a normal part of the development environment and do not impact the learning objectives.
PDFMiner is a specialized Python library focused on text extraction and layout analysis from PDF documents.
LangChain's PDFMinerLoader integrates with PDFMiner to parse PDF documents into LangChain Document objects.
from langchain_community.document_loaders import PDFMinerLoader# Create a PDFMiner loader instanceloader =PDFMinerLoader(FILE_PATH)# load datadocs = loader.load()# print the contents of the documentprint(docs[0].page_content[:300])
1
2
0
2
n
u
J
1
2
]
V
C
.
s
c
[
2
v
8
4
3
5
1
.
3
0
1
2
:
v
i
X
r
a
LayoutParser: A Unified Toolkit for Deep
Learning Based Document Image Analysis
Zejiang Shen1 ((cid:0)), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain
Lee4, Jacob Carlson3, and Weining Li5
1 Allen Institute for AI
s
This method allows you to parse the output HTML content through BeautifulSoup to get more structured and richer information about font size, page numbers, PDF header/footer, etc. which can help you semantically split the text into sections.
from langchain_community.document_loaders import PDFMinerPDFasHTMLLoader# create an instance of PDFMinerPDFasHTMLLoaderloader =PDFMinerPDFasHTMLLoader(FILE_PATH)# load the documentdocs = loader.load()# print the contents of the documentprint(docs[0].page_content[:300])
Page 1show_metadata(docs)[metadata] ['source'] [examples] source :./data/layout-parser-paper.pdffrom bs4 import BeautifulSoupsoup = BeautifulSoup(docs[0].page_content, "html.parser")# initialize HTML parsercontent = soup.find_all("div") # search for all div tagsimport recur_fs = Nonecur_text = ""snippets = [] # collect all snippets of the same font sizefor c in content: sp = c.find("span") if not sp: continue st = sp.get("style") if not st: continue fs = re.findall("font-size:(\d+)px", st) if not fs: continue fs = int(fs[0]) if not cur_fs: cur_fs = fs if fs == cur_fs: cur_text += c.text else: snippets.append((cur_text, cur_fs)) cur_fs = fs cur_text = c.textsnippets.append((cur_text, cur_fs))# Note: Possibility to add a strategy for removing duplicate snippets (since the header/footer of a PDF appears across multiple pages, it can be considered duplicate information when found)
from langchain_core.documents import Documentcur_idx = -1semantic_snippets = []# Assumption: headings have higher font size than their respective contentfor s in snippets: # if current snippet's font size > previous section's heading => it is a new heading if ( not semantic_snippets or s[1] > semantic_snippets[cur_idx].metadata["heading_font"] ): metadata = {"heading": s[0], "content_font": 0, "heading_font": s[1]} metadata.update(docs[0].metadata) semantic_snippets.append(Document(page_content="", metadata=metadata)) cur_idx += 1 continue # if current snippet's font size <= previous section's content => content belongs to the same section (one can also create if ( not semantic_snippets[cur_idx].metadata["content_font"] or s[1] <= semantic_snippets[cur_idx].metadata["content_font"] ): semantic_snippets[cur_idx].page_content += s[0] semantic_snippets[cur_idx].metadata["content_font"] = max( s[1], semantic_snippets[cur_idx].metadata["content_font"] ) continue # if current snippet's font size > previous section's content but less than previous section's heading than also make a new metadata = {"heading": s[0], "content_font": 0, "heading_font": s[1]} metadata.update(docs[0].metadata) semantic_snippets.append(Document(page_content="", metadata=metadata)) cur_idx += 1print(semantic_snippets[4])
page_content='Recently, various DL models and datasets have been developed for layout analysis tasks. The dhSegment [22] utilizes fully convolutional networks [20] for segmen- tation tasks on historical documents. Object detection-based methods like Faster R-CNN [28] and Mask R-CNN [12] are used for identifying document elements [38] and detecting tables [30, 26]. Most recently, Graph Neural Networks [29] have also been used in table detection [27]. However, these models are usually implemented individually and there is no unified framework to load and use such models. There has been a surge of interest in creating open-source tools for document image processing: a search of document image analysis in Github leads to 5M relevant code pieces 6; yet most of them rely on traditional rule-based methods or provide limited functionalities. The closest prior research to our work is the OCR-D project7, which also tries to build a complete toolkit for DIA. However, similar to the platform developed by Neudecker et al. [21], it is designed for analyzing historical documents, and provides no supports for recent DL models. The DocumentLayoutAnalysis project8 focuses on processing born-digital PDF documents via analyzing the stored PDF data. Repositories like DeepLayout9 and Detectron2-PubLayNet10 are individual deep learning models trained on layout analysis datasets without support for the full DIA pipeline. The Document Analysis and Exploitation (DAE) platform [15] and the DeepDIVA project [2] aim to improve the reproducibility of DIA methods (or DL models), yet they are not actively maintained. OCR engines like Tesseract [14], easyOCR11 and paddleOCR12 usually do not come with comprehensive functionalities for other DIA tasks like layout analysis. Recent years have also seen numerous efforts to create libraries for promoting reproducibility and reusability in the field of DL. Libraries like Dectectron2 [35], 6 The number shown is obtained by specifying the search type as ‘code’. 7 https://ocr-d.de/en/about 8 https://github.com/BobLd/DocumentLayoutAnalysis 9 https://github.com/leonlulu/DeepLayout 10 https://github.com/hpanwar08/detectron2 11 https://github.com/JaidedAI/EasyOCR 12 https://github.com/PaddlePaddle/PaddleOCR 4 Z. Shen et al. Fig. 1: The overall architecture of LayoutParser. For an input document image, the core LayoutParser library provides a set of off-the-shelf tools for layout detection, OCR, visualization, and storage, backed by a carefully designed layout data structure. LayoutParser also supports high level customization via efficient layout annotation and model training functions. These improve model accuracy on the target samples. The community platform enables the easy sharing of DIA models and whole digitization pipelines to promote reusability and reproducibility. A collection of detailed documentation, tutorials and exemplar projects make LayoutParser easy to learn and use. AllenNLP [8] and transformers [34] have provided the community with complete DL-based support for developing and deploying models for general computer vision and natural language processing problems. LayoutParser, on the other hand, specializes specifically in DIA tasks. LayoutParser is also equipped with a community platform inspired by established model hubs such as Torch Hub [23] and TensorFlow Hub [1]. It enables the sharing of pretrained models as well as full document processing pipelines that are unique to DIA tasks. There have been a variety of document data collections to facilitate the development of DL models. Some examples include PRImA [3](magazine layouts), PubLayNet [38](academic paper layouts), Table Bank [18](tables in academic papers), Newspaper Navigator Dataset [16, 17](newspaper figure layouts) and HJDataset [31](historical Japanese document layouts). A spectrum of models trained on these datasets are currently available in the LayoutParser model zoo to support different use cases. ' metadata={'heading': '2 Related Work\n', 'content_font': 9, 'heading_font': 11, 'source': './data/layout-parser-paper.pdf'}
PDFPlumberPDFPlumber is a PDF parsing library that excels at extracting text and tables from PDFs.LangChain's integrates with PDFPlumber to parse PDF documents into LangChain Document objects.Like PyMuPDF, the output document contains detailed metadata about the PDF and its pages,and returns one document per page.from langchain_community.document_loaders import PDFPlumberLoader# create a PDF document loader instanceloader = PDFPlumberLoader(FILE_PATH)# load the documentdocs = loader.load()# access the first document dataprint(docs[10].page_content[:300])LayoutParser: A Unified Toolkit for DL-Based DIA 11 focuses on precision, efficiency,and robustness. The target documents may have complicatedstructures,andmayrequiretrainingmultiplelayoutdetectionmodels to achieve the optimal accuracy. Light-weight pipelines are built for relatively simple documenshow_metadata(docs)[metadata] ['source', 'file_path', 'page', 'total_pages', 'Author', 'CreationDate', 'Creator', 'Keywords', 'ModDate', 'PTEX.Fullbanner', 'Producer', 'Subject', 'Title', 'Trapped'] [examples] source : ./data/layout-parser-paper.pdf file_path : ./data/layout-parser-paper.pdf page : 0 total_pages : 16 Author : CreationDate : D:20210622012710Z Creator : LaTeX with hyperref Keywords : ModDate : D:20210622012710Z PTEX.Fullbanner : This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2 Producer : pdfTeX-1.40.21 Subject : Title : Trapped : False
