1. 文档加载器
使用文档加载器从源加载数据作为 Document
。Document
是一段文本和相关元数据。例如,有用于加载简单的 .txt
文件、加载任何网页文本内容,甚至加载 YouTube 视频的转录文档加载器。
文档加载器提供一个 “load” 方法,用于从配置的源加载数据作为文档。它们还可以选择性地实现 “lazy load” 方法,以便将数据懒加载到内存中。
入门指南
最简单的加载器将文件读取为文本,并将其全部放入一个文档中。
from langchain_community.document_loaders import TextLoader
loader = TextLoader("./index.md")
loader.load()
[
Document(page_content='---\nsidebar_position: 0\n---\n# Document loaders\n\nUse document loaders to load data from a source as `Document`\'s. A `Document` is a piece of text\nand associated metadata. For example, there are document loaders for loading a simple `.txt` file, for loading the text\ncontents of any web page, or even for loading a transcript of a YouTube video.\n\nEvery document loader exposes two methods:\n1. "Load": load documents from the configured source\n2. "Load and split": load documents from the configured source and split them using the passed in text splitter\n\nThey optionally implement:\n\n3. "Lazy load": load documents into memory lazily\n', metadata={'source': '../docs/docs/modules/data_connection/document_loaders/index.md'})
]
2. 文件目录
这里介绍了如何加载目录中的所有文档。
在内部,默认情况下使用UnstructuredLoader。
from langchain_community.document_loaders import DirectoryLoader
我们可以使用glob
参数来控制加载哪些文件。请注意,这里不会加载.rst
文件或.html
文件。
loader = DirectoryLoader('../', glob="**/*.md")
docs = loader.load()
len(docs)
1
显示进度条
默认情况下不会显示进度条。要显示进度条,请安装tqdm
库(例如pip install tqdm
),并将show_progress
参数设置为True
。
loader = DirectoryLoader('../', glob="**/*.md", show_progress=True)
docs = loader.load()
Requirement already satisfied: tqdm in /Users/jon/.pyenv/versions/3.9.16/envs/microbiome-app/lib/python3.9/site-packages (4.65.0)
0it [00:00, ?it/s]
使用多线程
默认情况下,加载在一个线程中进行。为了利用多个线程,请将use_multithreading
标志设置为true。
loader = DirectoryLoader('../', glob="**/*.md", use_multithreading=True)
docs = loader.load()
更改加载器类
默认情况下使用UnstructuredLoader
类。但是,您可以很容易地更改加载器的类型。
from langchain_community.document_loaders import TextLoader
loader = DirectoryLoader('../', glob="**/*.md", loader_cls=TextLoader)
docs = loader.load()
len(docs)
1
如果需要加载Python源代码文件,请使用PythonLoader
。
from langchain_community.document_loaders import PythonLoader
loader = DirectoryLoader('../../../../../', glob="**/*.py", loader_cls=PythonLoader)
docs = loader.load()
len(docs)
691
使用TextLoader自动检测文件编码
在这个例子中,我们将看到一些在使用TextLoader
类从目录中加载大量任意文件时可能有用的策略。
首先,为了说明问题,让我们尝试加载具有任意编码的多个文本。
path = '../../../../../tests/integration_tests/examples'
loader = DirectoryLoader(path, glob="**/*.txt", loader_cls=TextLoader)
A. 默认行为
loader.load()
文件example-non-utf8.txt
使用了不同的编码,因此load()
函数失败,并显示了一个有用的消息,指示哪个文件解码失败。
使用TextLoader
的默认行为,任何文档加载失败都会导致整个加载过程失败,不会加载任何文档。
B. 静默失败
我们可以向DirectoryLoader
传递参数silent_errors
以跳过无法加载的文件并继续加载过程。
loader = DirectoryLoader(path, glob="**/*.txt", loader_cls=TextLoader, silent_errors=True)
docs = loader.load()
Error loading ../../../../../tests/integration_tests/examples/example-non-utf8.txt
doc_sources = [doc.metadata['source'] for doc in docs]
doc_sources
['../../../../../tests/integration_tests/examples/whatsapp_chat.txt',
'../../../../../tests/integration_tests/examples/example-utf8.txt']
C. 自动检测编码
我们还可以要求TextLoader
在失败之前自动检测文件编码,通过向加载器类传递autodetect_encoding
参数。
text_loader_kwargs={'autodetect_encoding': True}
loader = DirectoryLoader(path, glob="**/*.txt", loader_cls=TextLoader, loader_kwargs=text_loader_kwargs)
docs = loader.load()
doc_sources = [doc.metadata['source'] for doc in docs]
doc_sources
['../../../../../tests/integration_tests/examples/example-non-utf8.txt',
'../../../../../tests/integration_tests/examples/whatsapp_chat.txt',
'../../../../../tests/integration_tests/examples/example-utf8.txt']
3. CSV
CSV(逗号分隔值)文件是一种使用逗号分隔值的分隔文本文件。文件的每一行都是一个数据记录。每个记录由一个或多个字段组成,字段之间用逗号分隔。
使用每个文档一行的CSV数据加载。
from langchain_community.document_loaders.csv_loader import CSVLoader
loader = CSVLoader(file_path='./example_data/mlb_teams_2012.csv')
data = loader.load()
print(data)
[Document(page_content='Team: Nationals\n"Payroll (millions)": 81.34\n"Wins": 98', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 0}, lookup_index=0), Document(page_content='Team: Reds\n"Payroll (millions)": 82.20\n"Wins": 97', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 1}, lookup_index=0), Document(page_content='Team: Yankees\n"Payroll (millions)": 197.96\n"Wins": 95', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 2}, lookup_index=0), Document(page_content='Team: Giants\n"Payroll (millions)": 117.62\n"Wins": 94', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 3}, lookup_index=0), Document(page_content='Team: Braves\n"Payroll (millions)": 83.31\n"Wins": 94', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 4}, lookup_index=0), Document(page_content='Team: Athletics\n"Payroll (millions)": 55.37\n"Wins": 94', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 5}, lookup_index=0), Document(page_content='Team: Rangers\n"Payroll (millions)": 120.51\n"Wins": 93', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 6}, lookup_index=0), Document(page_content='Team: Orioles\n"Payroll (millions)": 81.43\n"Wins": 93', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 7}, lookup_index=0), Document(page_content='Team: Rays\n"Payroll (millions)": 64.17\n"Wins": 90', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 8}, lookup_index=0), Document(page_content='Team: Angels\n"Payroll (millions)": 154.49\n"Wins": 89', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 9}, lookup_index=0), Document(page_content='Team: Tigers\n"Payroll (millions)": 132.30\n"Wins": 88', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 10}, lookup_index=0), Document(page_content='Team: Cardinals\n"Payroll (millions)": 110.30\n"Wins": 88', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 11}, lookup_index=0), Document(page_content='Team: Dodgers\n"Payroll (millions)": 95.14\n"Wins": 86', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 12}, lookup_index=0), Document(page_content='Team: White Sox\n"Payroll (millions)": 96.92\n"Wins": 85', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 13}, lookup_index=0), Document(page_content='Team: Brewers\n"Payroll (millions)": 97.65\n"Wins": 83', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 14}, lookup_index=0), Document(page_content='Team: Phillies\n"Payroll (millions)": 174.54\n"Wins": 81', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 15}, lookup_index=0), Document(page_content='Team: Diamondbacks\n"Payroll (millions)": 74.28\n"Wins": 81', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 16}, lookup_index=0), Document(page_content='Team: Pirates\n"Payroll (millions)": 63.43\n"Wins": 79', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 17}, lookup_index=0), Document(page_content='Team: Padres\n"Payroll (millions)": 55.24\n"Wins": 76', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 18}, lookup_index=0), Document(page_content='Team: Mariners\n"Payroll (millions)": 81.97\n"Wins": 75', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 19}, lookup_index=0), Document(page_content='Team: Mets\n"Payroll (millions)": 93.35\n"Wins": 74', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 20}, lookup_index=0), Document(page_content='Team: Blue Jays\n"Payroll (millions)": 75.48\n"Wins": 73', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 21}, lookup_index=0), Document(page_content='Team: Royals\n"Payroll (millions)": 60.91\n"Wins": 72', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 22}, lookup_index=0), Document(page_content='Team: Marlins\n"Payroll (millions)": 118.07\n"Wins": 69', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 23}, lookup_index=0), Document(page_content='Team: Red Sox\n"Payroll (millions)": 173.18\n"Wins": 69', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 24}, lookup_index=0), Document(page_content='Team: Indians\n"Payroll (millions)": 78.43\n"Wins": 68', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 25}, lookup_index=0), Document(page_content='Team: Twins\n"Payroll (millions)": 94.08\n"Wins": 66', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 26}, lookup_index=0), Document(page_content='Team: Rockies\n"Payroll (millions)": 78.06\n"Wins": 64', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 27}, lookup_index=0), Document(page_content='Team: Cubs\n"Payroll (millions)": 88.19\n"Wins": 61', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 28}, lookup_index=0), Document(page_content='Team: Astros\n"Payroll (millions)": 60.65\n"Wins": 55', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 29}, lookup_index=0)]
自定义CSV解析和加载
有关支持的csv参数的更多信息,请参见csv模块文档。
loader = CSVLoader(file_path='./example_data/mlb_teams_2012.csv', csv_args={
'delimiter': ',',
'quotechar': '"',
'fieldnames': ['MLB Team', 'Payroll in millions', 'Wins']
})
data = loader.load()
print(data)
[Document(page_content='MLB Team: Team\nPayroll in millions: "Payroll (millions)"\nWins: "Wins"', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 0}, lookup_index=0), Document(page_content='MLB Team: Nationals\nPayroll in millions: 81.34\nWins: 98', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 1}, lookup_index=0), Document(page_content='MLB Team: Reds\nPayroll in millions: 82.20\nWins: 97', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 2}, lookup_index=0), Document(page_content='MLB Team: Yankees\nPayroll in millions: 197.96\nWins: 95', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 3}, lookup_index=0), Document(page_content='MLB Team: Giants\nPayroll in millions: 117.62\nWins: 94', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 4}, lookup_index=0), Document(page_content='MLB Team: Braves\nPayroll in millions: 83.31\nWins: 94', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 5}, lookup_index=0), Document(page_content='MLB Team: Athletics\nPayroll in millions: 55.37\nWins: 94', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 6}, lookup_index=0), Document(page_content='MLB Team: Rangers\nPayroll in millions: 120.51\nWins: 93', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 7}, lookup_index=0), Document(page_content='MLB Team: Orioles\nPayroll in millions: 81.43\nWins: 93', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 8}, lookup_index=0), Document(page_content='MLB Team: Rays\nPayroll in millions: 64.17\nWins: 90', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 9}, lookup_index=0), Document(page_content='MLB Team: Angels\nPayroll in millions: 154.49\nWins: 89', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 10}, lookup_index=0), Document(page_content='MLB Team: Tigers\nPayroll in millions: 132.30\nWins: 88', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 11}, lookup_index=0), Document(page_content='MLB Team: Cardinals\nPayroll in millions: 110.30\nWins: 88', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 12}, lookup_index=0), Document(page_content='MLB Team: Dodgers\nPayroll in millions: 95.14\nWins: 86', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 13}, lookup_index=0), Document(page_content='MLB Team: White Sox\nPayroll in millions: 96.92\nWins: 85', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 14}, lookup_index=0), Document(page_content='MLB Team: Brewers\nPayroll in millions: 97.65\nWins: 83', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 15}, lookup_index=0), Document(page_content='MLB Team: Phillies\nPayroll in millions: 174.54\nWins: 81', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 16}, lookup_index=0), Document(page_content='MLB Team: Diamondbacks\nPayroll in millions: 74.28\nWins: 81', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 17}, lookup_index=0), Document(page_content='MLB Team: Pirates\nPayroll in millions: 63.43\nWins: 79', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 18}, lookup_index=0), Document(page_content='MLB Team: Padres\nPayroll in millions: 55.24\nWins: 76', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 19}, lookup_index=0), Document(page_content='MLB Team: Mariners\nPayroll in millions: 81.97\nWins: 75', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 20}, lookup_index=0), Document(page_content='MLB Team: Mets\nPayroll in millions: 93.35\nWins: 74', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 21}, lookup_index=0), Document(page_content='MLB Team: Blue Jays\nPayroll in millions: 75.48\nWins: 73', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 22}, lookup_index=0), Document(page_content='MLB Team: Royals\nPayroll in millions: 60.91\nWins: 72', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 23}, lookup_index=0), Document(page_content='MLB Team: Marlins\nPayroll in millions: 118.07\nWins: 69', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 24}, lookup_index=0), Document(page_content='MLB Team: Red Sox\nPayroll in millions: 173.18\nWins: 69', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 25}, lookup_index=0), Document(page_content='MLB Team: Indians\nPayroll in millions: 78.43\nWins: 68', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 26}, lookup_index=0), Document(page_content='MLB Team: Twins\nPayroll in millions: 94.08\nWins: 66', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 27}, lookup_index=0), Document(page_content='MLB Team: Rockies\nPayroll in millions: 78.06\nWins: 64', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 28}, lookup_index=0), Document(page_content='MLB Team: Cubs\nPayroll in millions: 88.19\nWins: 61', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 29}, lookup_index=0), Document(page_content='MLB Team: Astros\nPayroll in millions: 60.65\nWins: 55', lookup_str='', metadata={'source': './example_data/mlb_teams_2012.csv', 'row': 30}, lookup_index=0)]
指定用于识别文档来源的列
使用 source_column
参数来指定从每一行创建的文档的来源。否则,将使用 file_path
作为从 CSV 文件创建的所有文档的来源。
当使用从 CSV 文件加载的文档用于回答问题的链式操作时,这将非常有用。
loader = CSVLoader(file_path='./example_data/mlb_teams_2012.csv', source_column="Team")
data = loader.load()
print(data)
[Document(page_content='Team: Nationals\n"Payroll (millions)": 81.34\n"Wins": 98', lookup_str='', metadata={'source': 'Nationals', 'row': 0}, lookup_index=0), Document(page_content='Team: Reds\n"Payroll (millions)": 82.20\n"Wins": 97', lookup_str='', metadata={'source': 'Reds', 'row': 1}, lookup_index=0), Document(page_content='Team: Yankees\n"Payroll (millions)": 197.96\n"Wins": 95', lookup_str='', metadata={'source': 'Yankees', 'row': 2}, lookup_index=0), Document(page_content='Team: Giants\n"Payroll (millions)": 117.62\n"Wins": 94', lookup_str='', metadata={'source': 'Giants', 'row': 3}, lookup_index=0), Document(page_content='Team: Braves\n"Payroll (millions)": 83.31\n"Wins": 94', lookup_str='', metadata={'source': 'Braves', 'row': 4}, lookup_index=0), Document(page_content='Team: Athletics\n"Payroll (millions)": 55.37\n"Wins": 94', lookup_str='', metadata={'source': 'Athletics', 'row': 5}, lookup_index=0), Document(page_content='Team: Rangers\n"Payroll (millions)": 120.51\n"Wins": 93', lookup_str='', metadata={'source': 'Rangers', 'row': 6}, lookup_index=0), Document(page_content='Team: Orioles\n"Payroll (millions)": 81.43\n"Wins": 93', lookup_str='', metadata={'source': 'Orioles', 'row': 7}, lookup_index=0), Document(page_content='Team: Rays\n"Payroll (millions)": 64.17\n"Wins": 90', lookup_str='', metadata={'source': 'Rays', 'row': 8}, lookup_index=0), Document(page_content='Team: Angels\n"Payroll (millions)": 154.49\n"Wins": 89', lookup_str='', metadata={'source': 'Angels', 'row': 9}, lookup_index=0), Document(page_content='Team: Tigers\n"Payroll (millions)": 132.30\n"Wins": 88', lookup_str='', metadata={'source': 'Tigers', 'row': 10}, lookup_index=0), Document(page_content='Team: Cardinals\n"Payroll (millions)": 110.30\n"Wins": 88', lookup_str='', metadata={'source': 'Cardinals', 'row': 11}, lookup_index=0), Document(page_content='Team: Dodgers\n"Payroll (millions)": 95.14\n"Wins": 86', lookup_str='', metadata={'source': 'Dodgers', 'row': 12}, lookup_index=0), Document(page_content='Team: White Sox\n"Payroll (millions)": 96.92\n"Wins": 85', lookup_str='', metadata={'source': 'White Sox', 'row': 13}, lookup_index=0), Document(page_content='Team: Brewers\n"Payroll (millions)": 97.65\n"Wins": 83', lookup_str='', metadata={'source': 'Brewers', 'row': 14}, lookup_index=0), Document(page_content='Team: Phillies\n"Payroll (millions)": 174.54\n"Wins": 81', lookup_str='', metadata={'source': 'Phillies', 'row': 15}, lookup_index=0), Document(page_content='Team: Diamondbacks\n"Payroll (millions)": 74.28\n"Wins": 81', lookup_str='', metadata={'source': 'Diamondbacks', 'row': 16}, lookup_index=0), Document(page_content='Team: Pirates\n"Payroll (millions)": 63.43\n"Wins": 79', lookup_str='', metadata={'source': 'Pirates', 'row': 17}, lookup_index=0), Document(page_content='Team: Padres\n"Payroll (millions)": 55.24\n"Wins": 76', lookup_str='', metadata={'source': 'Padres', 'row': 18}, lookup_index=0), Document(page_content='Team: Mariners\n"Payroll (millions)": 81.97\n"Wins": 75', lookup_str='', metadata={'source': 'Mariners', 'row': 19}, lookup_index=0), Document(page_content='Team: Mets\n"Payroll (millions)": 93.35\n"Wins": 74', lookup_str='', metadata={'source': 'Mets', 'row': 20}, lookup_index=0), Document(page_content='Team: Blue Jays\n"Payroll (millions)": 75.48\n"Wins": 73', lookup_str='', metadata={'source': 'Blue Jays', 'row': 21}, lookup_index=0), Document(page_content='Team: Royals\n"Payroll (millions)": 60.91\n"Wins": 72', lookup_str='', metadata={'source': 'Royals', 'row': 22}, lookup_index=0), Document(page_content='Team: Marlins\n"Payroll (millions)": 118.07\n"Wins": 69', lookup_str='', metadata={'source': 'Marlins', 'row': 23}, lookup_index=0), Document(page_content='Team: Red Sox\n"Payroll (millions)": 173.18\n"Wins": 69', lookup_str='', metadata={'source': 'Red Sox', 'row': 24}, lookup_index=0), Document(page_content='Team: Indians\n"Payroll (millions)": 78.43\n"Wins": 68', lookup_str='', metadata={'source': 'Indians', 'row': 25}, lookup_index=0), Document(page_content='Team: Twins\n"Payroll (millions)": 94.08\n"Wins": 66', lookup_str='', metadata={'source': 'Twins', 'row': 26}, lookup_index=0), Document(page_content='Team: Rockies\n"Payroll (millions)": 78.06\n"Wins": 64', lookup_str='', metadata={'source': 'Rockies', 'row': 27}, lookup_index=0), Document(page_content='Team: Cubs\n"Payroll (millions)": 88.19\n"Wins": 61', lookup_str='', metadata={'source': 'Cubs', 'row': 28}, lookup_index=0), Document(page_content='Team: Astros\n"Payroll (millions)": 60.65\n"Wins": 55', lookup_str='', metadata={'source': 'Astros', 'row': 29}, lookup_index=0)]
4. Markdown
Markdown 是一种轻量级标记语言,可通过纯文本编辑器创建格式化文本。
这里介绍了如何将 Markdown
文档加载到我们可以在下游使用的文档格式中。
# !pip install unstructured > /dev/null
from langchain_community.document_loaders import UnstructuredMarkdownLoader
markdown_path = "../../../../../README.md"
loader = UnstructuredMarkdownLoader(markdown_path)
data = loader.load()
data
[Document(page_content="ð\x9f¦\x9cï¸\x8fð\x9f”\x97 LangChain\n\nâ\x9a¡ Building applications with LLMs through composability â\x9a¡\n\nLooking for the JS/TS version? Check out LangChain.js.\n\nProduction Support: As you move your LangChains into production, we'd love to offer more comprehensive support.\nPlease fill out this form and we'll set up a dedicated support Slack channel.\n\nQuick Install\n\npip install langchain\nor\nconda install langchain -c conda-forge\n\nð\x9f¤” What is this?\n\nLarge language models (LLMs) are emerging as a transformative technology, enabling developers to build applications that they previously could not. However, using these LLMs in isolation is often insufficient for creating a truly powerful app - the real power comes when you can combine them with other sources of computation or knowledge.\n\nThis library aims to assist in the development of those types of applications. Common examples of these applications include:\n\nâ\x9d“ Question Answering over specific documents\n\nDocumentation\n\nEnd-to-end Example: Question Answering over Notion Database\n\nð\x9f’¬ Chatbots\n\nDocumentation\n\nEnd-to-end Example: Chat-LangChain\n\nð\x9f¤\x96 Agents\n\nDocumentation\n\nEnd-to-end Example: GPT+WolframAlpha\n\nð\x9f“\x96 Documentation\n\nPlease see here for full documentation on:\n\nGetting started (installation, setting up the environment, simple examples)\n\nHow-To examples (demos, integrations, helper functions)\n\nReference (full API docs)\n\nResources (high-level explanation of core concepts)\n\nð\x9f\x9a\x80 What can this help with?\n\nThere are six main areas that LangChain is designed to help with.\nThese are, in increasing order of complexity:\n\nð\x9f“\x83 LLMs and Prompts:\n\nThis includes prompt management, prompt optimization, a generic interface for all LLMs, and common utilities for working with LLMs.\n\nð\x9f”\x97 Chains:\n\nChains go beyond a single LLM call and involve sequences of calls (whether to an LLM or a different utility). LangChain provides a standard interface for chains, lots of integrations with other tools, and end-to-end chains for common applications.\n\nð\x9f“\x9a Data Augmented Generation:\n\nData Augmented Generation involves specific types of chains that first interact with an external data source to fetch data for use in the generation step. Examples include summarization of long pieces of text and question/answering over specific data sources.\n\nð\x9f¤\x96 Agents:\n\nAgents involve an LLM making decisions about which Actions to take, taking that Action, seeing an Observation, and repeating that until done. LangChain provides a standard interface for agents, a selection of agents to choose from, and examples of end-to-end agents.\n\nð\x9f§\xa0 Memory:\n\nMemory refers to persisting state between calls of a chain/agent. LangChain provides a standard interface for memory, a collection of memory implementations, and examples of chains/agents that use memory.\n\nð\x9f§\x90 Evaluation:\n\n[BETA] Generative models are notoriously hard to evaluate with traditional metrics. One new way of evaluating them is using language models themselves to do the evaluation. LangChain provides some prompts/chains for assisting in this.\n\nFor more information on these concepts, please see our full documentation.\n\nð\x9f’\x81 Contributing\n\nAs an open-source project in a rapidly developing field, we are extremely open to contributions, whether it be in the form of a new feature, improved infrastructure, or better documentation.\n\nFor detailed information on how to contribute, see here.", metadata={'source': '../../../../../README.md'})]
保留元素
在幕后,Unstructured 为不同的文本块创建不同的 “元素”。默认情况下,我们将它们组合在一起,但您可以通过指定 mode="elements"
轻松保留该分隔。
loader = UnstructuredMarkdownLoader(markdown_path, mode="elements")
data = loader.load()
data[0]
Document(page_content='ð\x9f¦\x9cï¸\x8fð\x9f”\x97 LangChain', metadata={'source': '../../../../../README.md', 'page_number': 1, 'category': 'Title'})
5. JSON
JSON(JavaScript对象表示法)是一种开放标准文件格式和数据交换格式,使用人类可读的文本来存储和传输由属性-值对和数组(或其他可序列化值)组成的数据对象。
JSON Lines是一种文件格式,其中每一行都是一个有效的JSON值。
JSONLoader
使用指定的 jq模式 来解析JSON文件。它使用jq
Python包。
查看这个 手册 以获取有关jq
语法的详细文档。
#!pip install jq
from langchain_community.document_loaders import JSONLoader
import json
from pathlib import Path
from pprint import pprint
file_path='./example_data/facebook_chat.json'
data = json.loads(Path(file_path).read_text())
pprint(data)
{'image': {'creation_timestamp': 1675549016, 'uri': 'image_of_the_chat.jpg'},
'is_still_participant': True,
'joinable_mode': {'link': '', 'mode': 1},
'magic_words': [],
'messages': [{'content': 'Bye!',
'sender_name': 'User 2',
'timestamp_ms': 1675597571851},
{'content': 'Oh no worries! Bye',
'sender_name': 'User 1',
'timestamp_ms': 1675597435669},
{'content': 'No Im sorry it was my mistake, the blue one is not '
'for sale',
'sender_name': 'User 2',
'timestamp_ms': 1675596277579},
{'content': 'I thought you were selling the blue one!',
'sender_name': 'User 1',
'timestamp_ms': 1675595140251},
{'content': 'Im not interested in this bag. Im interested in the '
'blue one!',
'sender_name': 'User 1',
'timestamp_ms': 1675595109305},
{'content': 'Here is $129',
'sender_name': 'User 2',
'timestamp_ms': 1675595068468},
{'photos': [{'creation_timestamp': 1675595059,
'uri': 'url_of_some_picture.jpg'}],
'sender_name': 'User 2',
'timestamp_ms': 1675595060730},
{'content': 'Online is at least $100',
'sender_name': 'User 2',
'timestamp_ms': 1675595045152},
{'content': 'How much do you want?',
'sender_name': 'User 1',
'timestamp_ms': 1675594799696},
{'content': 'Goodmorning! $50 is too low.',
'sender_name': 'User 2',
'timestamp_ms': 1675577876645},
{'content': 'Hi! Im interested in your bag. Im offering $50. Let '
'me know if you are interested. Thanks!',
'sender_name': 'User 1',
'timestamp_ms': 1675549022673}],
'participants': [{'name': 'User 1'}, {'name': 'User 2'}],
'thread_path': 'inbox/User 1 and User 2 chat',
'title': 'User 1 and User 2 chat'}
使用 JSONLoader
假设我们有兴趣提取JSON数据中 messages
键下 content
字段的值。可以通过 JSONLoader
轻松实现如下。
JSON文件
loader = JSONLoader(
file_path='./example_data/facebook_chat.json',
jq_schema='.messages[].content',
text_content=False)
data = loader.load()
pprint(data)
[Document(page_content='Bye!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 1}),
Document(page_content='Oh no worries! Bye', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 2}),
Document(page_content='No Im sorry it was my mistake, the blue one is not for sale', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 3}),
Document(page_content='I thought you were selling the blue one!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 4}),
Document(page_content='Im not interested in this bag. Im interested in the blue one!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 5}),
Document(page_content='Here is $129', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 6}),
Document(page_content='', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 7}),
Document(page_content='Online is at least $100', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 8}),
Document(page_content='How much do you want?', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 9}),
Document(page_content='Goodmorning! $50 is too low.', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 10}),
Document(page_content='Hi! Im interested in your bag. Im offering $50. Let me know if you are interested. Thanks!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 11})]
JSON Lines文件
如果要从JSON Lines文件加载文档,可以传递 json_lines=True
,并指定 jq_schema
以从单个JSON对象中提取 page_content
。
file_path = './example_data/facebook_chat_messages.jsonl'
pprint(Path(file_path).read_text())
('{"sender_name": "User 2", "timestamp_ms": 1675597571851, "content": "Bye!"}\n'
'{"sender_name": "User 1", "timestamp_ms": 1675597435669, "content": "Oh no '
'worries! Bye"}\n'
'{"sender_name": "User 2", "timestamp_ms": 1675596277579, "content": "No Im '
'sorry it was my mistake, the blue one is not for sale"}\n')
loader = JSONLoader(
file_path='./example_data/facebook_chat_messages.jsonl',
jq_schema='.content',
text_content=False,
json_lines=True)
data = loader.load()
pprint(data)
[Document(page_content='Bye!', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat_messages.jsonl', 'seq_num': 1}),
Document(page_content='Oh no worries! Bye', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat_messages.jsonl', 'seq_num': 2}),
Document(page_content='No Im sorry it was my mistake, the blue one is not for sale', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat_messages.jsonl', 'seq_num': 3})]
另一个选项是设置 jq_schema='.'
并提供 content_key
:
loader = JSONLoader(
file_path='./example_data/facebook_chat_messages.jsonl',
jq_schema='.',
content_key='sender_name',
json_lines=True)
data = loader.load()
pprint(data)
[Document(page_content='User 2', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat_messages.jsonl', 'seq_num': 1}),
Document(page_content='User 1', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat_messages.jsonl', 'seq_num': 2}),
Document(page_content='User 2', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat_messages.jsonl', 'seq_num': 3})]
具有jq模式 content_key
的JSON文件
要使用jq模式内的 content_key
从JSON文件加载文档,请设置 is_content_key_jq_parsable=True
。
确保 content_key
兼容并可以使用jq模式进行解析。
file_path = './sample.json'
pprint(Path(file_path).read_text())
{"data": [
{"attributes": {
"message": "message1",
"tags": [
"tag1"]},
"id": "1"},
{"attributes": {
"message": "message2",
"tags": [
"tag2"]},
"id": "2"}]}
loader = JSONLoader(
file_path=file_path,
jq_schema=".data[]",
content_key=".attributes.message",
is_content_key_jq_parsable=True,
)
data = loader.load()
pprint(data)
[Document(page_content='message1', metadata={'source': '/path/to/sample.json', 'seq_num': 1}),
Document(page_content='message2', metadata={'source': '/path/to/sample.json', 'seq_num': 2})]
提取元数据
通常,我们希望将JSON文件中的元数据包含在从内容创建的文档中。
以下演示了如何使用 JSONLoader
提取元数据。
在当前示例中,我们需要告诉加载器迭代messages
字段中的记录。然后,jq_schema
应为:
.messages[]
这使我们能够将记录(字典)传递给必须实现的metadata_func
。metadata_func
负责识别记录中应包含在最终Document
对象中存储的元数据的哪些信息。
此外,现在我们必须在加载器中明确指定通过content_key
参数从记录中提取page_content
值的键。
# 定义元数据提取函数。
def metadata_func(record: dict, metadata: dict) -> dict:
metadata["sender_name"] = record.get("sender_name")
metadata["timestamp_ms"] = record.get("timestamp_ms")
return metadata
loader = JSONLoader(
file_path='./example_data/facebook_chat.json',
jq_schema='.messages[]',
content_key="content",
metadata_func=metadata_func
)
data = loader.load()
pprint(data)
[Document(page_content='再见!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 1, 'sender_name': 'User 2', 'timestamp_ms': 1675597571851}),
Document(page_content='哦,没关系!再见', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 2, 'sender_name': 'User 1', 'timestamp_ms': 1675597435669}),
Document(page_content='不好意思,是我的错,蓝色的不出售', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 3, 'sender_name': 'User 2', 'timestamp_ms': 1675596277579}),
Document(page_content='我以为你在卖蓝色的那个!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 4, 'sender_name': 'User 1', 'timestamp_ms': 1675595140251}),
Document(page_content='我对这个包不感兴趣。我对蓝色的感兴趣!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 5, 'sender_name': 'User 1', 'timestamp_ms': 1675595109305}),
Document(page_content='这里是129美元', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 6, 'sender_name': 'User 2', 'timestamp_ms': 1675595068468}),
Document(page_content='', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 7, 'sender_name': 'User 2', 'timestamp_ms': 1675595060730}),
Document(page_content='网上至少100美元', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 8, 'sender_name': 'User 2', 'timestamp_ms': 1675595045152}),
Document(page_content='你想要多少?', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 9, 'sender_name': 'User 1', 'timestamp_ms': 1675594799696}),
Document(page_content='早上好!50美元太少了。', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 10, 'sender_name': 'User 2', 'timestamp_ms': 1675577876645}),
Document(page_content='嗨!我对你的包感兴趣。我出价50美元。如果你有兴趣,请告诉我。谢谢!', metadata={'source': '/Users/avsolatorio/WBG/langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 11, 'sender_name': 'User 1', 'timestamp_ms': 1675549022673})]
现在,您会看到文档包含与我们提取的内容相关的元数据。
metadata_func
如上所示,metadata_func
接受JSONLoader
生成的默认元数据。这使用户完全控制元数据的格式。
例如,默认元数据包含source
和seq_num
键。但是,JSON数据中也可能包含这些键。用户可以利用metadata_func
重命名默认键并使用JSON数据中的键。
下面的示例显示了如何修改source
,使其仅包含相对于langchain
目录的文件源信息。
# 定义元数据提取函数。
def metadata_func(record: dict, metadata: dict) -> dict:
metadata["sender_name"] = record.get("sender_name")
metadata["timestamp_ms"] = record.get("timestamp_ms")
if "source" in metadata:
source = metadata["source"].split("/")
source = source[source.index("langchain"):]
metadata["source"] = "/".join(source)
return metadata
loader = JSONLoader(
file_path='./example_data/facebook_chat.json',
jq_schema='.messages[]',
content_key="content",
metadata_func=metadata_func
)
data = loader.load()
pprint(data)
[Document(page_content='再见!', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 1, 'sender_name': 'User 2', 'timestamp_ms': 1675597571851}),
Document(page_content='哦,没关系!再见', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 2, 'sender_name': 'User 1', 'timestamp_ms': 1675597435669}),
Document(page_content='不好意思,是我的错,蓝色的不出售', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 3, 'sender_name': 'User 2', 'timestamp_ms': 1675596277579}),
Document(page_content='我以为你在卖蓝色的那个!', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 4, 'sender_name': 'User 1', 'timestamp_ms': 1675595140251}),
Document(page_content='我对这个包不感兴趣。我对蓝色的感兴趣!', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 5, 'sender_name': 'User 1', 'timestamp_ms': 1675595109305}),
Document(page_content='这里是129美元', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 6, 'sender_name': 'User 2', 'timestamp_ms': 1675595068468}),
Document(page_content='', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 7, 'sender_name': 'User 2', 'timestamp_ms': 1675595060730}),
Document(page_content='网上至少100美元', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 8, 'sender_name': 'User 2', 'timestamp_ms': 1675595045152}),
Document(page_content='你想要多少?', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 9, 'sender_name': 'User 1', 'timestamp_ms': 1675594799696}),
Document(page_content='早上好!50美元太少了。', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 10, 'sender_name': 'User 2', 'timestamp_ms': 1675577876645}),
Document(page_content='嗨!我对你的包感兴趣。我出价50美元。如果你有兴趣,请告诉我。谢谢!', metadata={'source': 'langchain/docs/modules/indexes/document_loaders/examples/example_data/facebook_chat.json', 'seq_num': 11, 'sender_name': 'User 1', 'timestamp_ms': 1675549022673})]
带有jq schema的常见JSON结构
以下列表提供了用户可以根据结构从JSON数据中提取内容的可能jq_schema
的参考。
JSON -> [{"text": ...}, {"text": ...}, {"text": ...}]
jq_schema -> ".[].text"
JSON -> {"key": [{"text": ...}, {"text": ...}, {"text": ...}]}
jq_schema -> ".key[].text"
JSON -> ["...", "...", "..."]
jq_schema -> ".[]"
6. PDF
便携式文档格式(PDF)标准化为ISO 32000,是由Adobe于1992年开发的文件格式,用于以与应用软件、硬件和操作系统无关的方式呈现文档,包括文本格式和图像。
这里介绍了如何将PDF
文档加载到我们下游使用的文档格式中。
使用 PyPDF
使用pypdf
将PDF加载到文档数组中,其中每个文档包含页面内容和带有page
编号的元数据。
pip install pypdf
from langchain_community.document_loaders import PyPDFLoader
loader = PyPDFLoader("example_data/layout-parser-paper.pdf")
pages = loader.load_and_split()
pages[0]
Document(page_content='LayoutParser : A Uni\x0ced Toolkit for Deep\nLearning Based Document Image Analysis\nZejiang Shen1( \x00), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain\nLee4, Jacob Carlson3, and Weining Li5\n1Allen Institute for AI\nshannons@allenai.org\n2Brown University\nruochen zhang@brown.edu\n3Harvard University\nfmelissadell,jacob carlson g@fas.harvard.edu\n4University of Washington\nbcgl@cs.washington.edu\n5University of Waterloo\nw422li@uwaterloo.ca\nAbstract. Recent advances in document image analysis (DIA) have been\nprimarily driven by the application of neural networks. Ideally, research\noutcomes could be easily deployed in production and extended for further\ninvestigation. However, various factors like loosely organized codebases\nand sophisticated model con\x0cgurations complicate the easy reuse of im-\nportant innovations by a wide audience. Though there have been on-going\ne\x0borts to improve reusability and simplify deep learning (DL) model\ndevelopment in disciplines like natural language processing and computer\nvision, none of them are optimized for challenges in the domain of DIA.\nThis represents a major gap in the existing toolkit, as DIA is central to\nacademic research across a wide range of disciplines in the social sciences\nand humanities. This paper introduces LayoutParser , an open-source\nlibrary for streamlining the usage of DL in DIA research and applica-\ntions. The core LayoutParser library comes with a set of simple and\nintuitive interfaces for applying and customizing DL models for layout de-\ntection, character recognition, and many other document processing tasks.\nTo promote extensibility, LayoutParser also incorporates a community\nplatform for sharing both pre-trained models and full document digiti-\nzation pipelines. We demonstrate that LayoutParser is helpful for both\nlightweight and large-scale digitization pipelines in real-word use cases.\nThe library is publicly available at https://layout-parser.github.io .\nKeywords: Document Image Analysis ·Deep Learning ·Layout Analysis\n·Character Recognition ·Open Source library ·Toolkit.\n1 Introduction\nDeep Learning(DL)-based approaches are the state-of-the-art for a wide range of\ndocument image analysis (DIA) tasks including document image classi\x0ccation [ 11,arXiv:2103.15348v2 [cs.CV] 21 Jun 2021', metadata={'source': 'example_data/layout-parser-paper.pdf', 'page': 0})
这种方法的优点是可以通过页码检索文档。
我们想要使用OpenAIEmbeddings
,因此必须获取OpenAI API密钥。
import os
import getpass
os.environ['OPENAI_API_KEY'] = getpass.getpass('OpenAI API Key:')
OpenAI API Key: ········
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings
faiss_index = FAISS.from_documents(pages, OpenAIEmbeddings())
docs = faiss_index.similarity_search("How will the community be engaged?", k=2)
for doc in docs:
print(str(doc.metadata["page"]) + ":", doc.page_content[:300])
9: 10 Z. Shen et al.
Fig. 4: Illustration of (a) the original historical Japanese document with layout
detection results and (b) a recreated version of the document image that achieves
much better character recognition recall. The reorganization algorithm rearranges
the tokens based on the their detect
3: 4 Z. Shen et al.
Efficient Data AnnotationC u s t o m i z e d M o d e l T r a i n i n gModel Cust omizationDI A Model HubDI A Pipeline SharingCommunity PlatformLa y out Detection ModelsDocument Images
T h e C o r e L a y o u t P a r s e r L i b r a r yOCR ModuleSt or age & VisualizationLa y ou
提取图像
使用rapidocr-onnxruntime
包,我们可以将图像提取为文本:
pip install rapidocr-onnxruntime
loader = PyPDFLoader("https://arxiv.org/pdf/2103.15348.pdf", extract_images=True)
pages = loader.load()
pages[4].page_content
'LayoutParser : A Unified Toolkit for DL-Based DIA 5\nTable 1: Current layout detection models in the LayoutParser model zoo\nDataset Base Model1Large Model Notes\nPubLayNet [38] F / M M Layouts of modern scientific documents\nPRImA [3] M - Layouts of scanned modern magazines and scientific reports\nNewspaper [17] F - Layouts of scanned US newspapers from the 20th century\nTableBank [18] F F Table region on modern scientific and business document\nHJDataset [31] F / M - Layouts of history Japanese documents\n1For each dataset, we train several models of different sizes for different needs (the trade-off between accuracy\nvs. computational cost). For “base model” and “large model”, we refer to using the ResNet 50 or ResNet 101\nbackbones [ 13], respectively. One can train models of different architectures, like Faster R-CNN [ 28] (F) and Mask\nR-CNN [ 12] (M). For example, an F in the Large Model column indicates it has a Faster R-CNN model trained\nusing the ResNet 101 backbone. The platform is maintained and a number of additions will be made to the model\nzoo in coming months.\nlayout data structures , which are optimized for efficiency and versatility. 3) When\nnecessary, users can employ existing or customized OCR models via the unified\nAPI provided in the OCR module . 4)LayoutParser comes with a set of utility\nfunctions for the visualization and storage of the layout data. 5) LayoutParser\nis also highly customizable, via its integration with functions for layout data\nannotation and model training . We now provide detailed descriptions for each\ncomponent.\n3.1 Layout Detection Models\nInLayoutParser , a layout model takes a document image as an input and\ngenerates a list of rectangular boxes for the target content regions. Different\nfrom traditional methods, it relies on deep convolutional neural networks rather\nthan manually curated rules to identify content regions. It is formulated as an\nobject detection problem and state-of-the-art models like Faster R-CNN [ 28] and\nMask R-CNN [ 12] are used. This yields prediction results of high accuracy and\nmakes it possible to build a concise, generalized interface for layout detection.\nLayoutParser , built upon Detectron2 [ 35], provides a minimal API that can\nperform layout detection with only four lines of code in Python:\n1import layoutparser as lp\n2image = cv2. imread (" image_file ") # load images\n3model = lp. Detectron2LayoutModel (\n4 "lp :// PubLayNet / faster_rcnn_R_50_FPN_3x / config ")\n5layout = model . detect ( image )\nLayoutParser provides a wealth of pre-trained model weights using various\ndatasets covering different languages, time periods, and document types. Due to\ndomain shift [ 7], the prediction performance can notably drop when models are ap-\nplied to target samples that are significantly different from the training dataset. As\ndocument structures and layouts vary greatly in different domains, it is important\nto select models trained on a dataset similar to the test samples. A semantic syntax\nis used for initializing the model weights in LayoutParser , using both the dataset\nname and model name lp://<dataset-name>/<model-architecture-name> .'
使用 MathPix
受Daniel Gross的https://gist.github.com/danielgross/3ab4104e14faccc12b49200843adab21启发
from langchain_community.document_loaders import MathpixPDFLoader
loader = MathpixPDFLoader("example_data/layout-parser-paper.pdf")
data = loader.load()
使用 Unstructured
from langchain_community.document_loaders import UnstructuredPDFLoader
loader = UnstructuredPDFLoader("example_data/layout-parser-paper.pdf")
data = loader.load()
保留元素
在幕后,Unstructured为不同的文本块创建不同的“元素”。默认情况下,我们将它们合并在一起,但您可以通过指定mode="elements"
来保持分离。
loader = UnstructuredPDFLoader("example_data/layout-parser-paper.pdf", mode="elements")
data = loader.load()
data[0]
Document(page_content='LayoutParser: A Unified Toolkit for Deep\nLearning Based Document Image Analysis\nZejiang Shen1 (�), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain\nLee4, Jacob Carlson3, and Weining Li5\n1 Allen Institute for AI\nshannons@allenai.org\n2 Brown University\nruochen zhang@brown.edu\n3 Harvard University\n{melissadell,jacob carlson}@fas.harvard.edu\n4 University of Washington\nbcgl@cs.washington.edu\n5 University of Waterloo\nw422li@uwaterloo.ca\nAbstract. Recent advances in document image analysis (DIA) have been\nprimarily driven by the application of neural networks. Ideally, research\noutcomes could be easily deployed in production and extended for further\ninvestigation. However, various factors like loosely organized codebases\nand sophisticated model configurations complicate the easy reuse of im-\nportant innovations by a wide audience. Though there have been on-going\nefforts to improve reusability and simplify deep learning (DL) model\ndevelopment in disciplines like natural language processing and computer\nvision, none of them are optimized for challenges in the domain of DIA.\nThis represents a major gap in the existing toolkit, as DIA is central to\nacademic research across a wide range of disciplines in the social sciences\nand humanities. This paper introduces LayoutParser, an open-source\nlibrary for streamlining the usage of DL in DIA research and applica-\ntions. The core LayoutParser library comes with a set of simple and\nintuitive interfaces for applying and customizing DL models for layout de-\ntection, character recognition, and many other document processing tasks.\nTo promote extensibility, LayoutParser also incorporates a community\nplatform for sharing both pre-trained models and full document digiti-\nzation pipelines. We demonstrate that LayoutParser is helpful for both\nlightweight and large-scale digitization pipelines in real-word use cases.\nThe library is publicly available at https://layout-parser.github.io.\nKeywords: Document Image Analysis · Deep Learning · Layout Analysis\n· Character Recognition · Open Source library · Toolkit.\n1\nIntroduction\nDeep Learning(DL)-based approaches are the state-of-the-art for a wide range of\ndocument image analysis (DIA) tasks including document image classification [11,\narXiv:2103.15348v2 [cs.CV] 21 Jun 2021\n', lookup_str='', metadata={'file_path': 'example_data/layout-parser-paper.pdf', 'page_number': 1, 'total_pages': 16, 'format': 'PDF 1.5', 'title': '', 'author': '', 'subject': '', 'keywords': '', 'creator': 'LaTeX with hyperref', 'producer': 'pdfTeX-1.40.21', 'creationDate': 'D:20210622012710Z', 'modDate': 'D:20210622012710Z', 'trapped': '', 'encryption': None}, lookup_index=0)
使用Unstructured获取远程PDF文件
本文介绍了如何将在线PDF文件加载为我们可以在下游使用的文档格式。这可以用于各种在线PDF网站,如https://open.umn.edu/opentextbooks/textbooks/和https://arxiv.org/archive/
注意:所有其他PDF加载器也可以用于获取远程PDF文件,但OnlinePDFLoader
是一个遗留函数,专门与UnstructuredPDFLoader
一起使用。
from langchain_community.document_loaders import OnlinePDFLoader
loader = OnlinePDFLoader("https://arxiv.org/pdf/2302.03803.pdf")
data = loader.load()
print(data)
[Document(page_content='A WEAK ( k, k ) -LEFSCHETZ THEOREM FOR PROJECTIVE TORIC ORBIFOLDS\n\nWilliam D. Montoya\n\nInstituto de Matem´atica, Estat´ıstica e Computa¸c˜ao Cient´ıfica,\n\nIn [3] we proved that, under suitable conditions, on a very general codimension s quasi- smooth intersection subvariety X in a projective toric orbifold P d Σ with d + s = 2 ( k + 1 ) the Hodge conjecture holds, that is, every ( p, p ) -cohomology class, under the Poincar´e duality is a rational linear combination of fundamental classes of algebraic subvarieties of X . The proof of the above-mentioned result relies, for p ≠ d + 1 − s , on a Lefschetz\n\nKeywords: (1,1)- Lefschetz theorem, Hodge conjecture, toric varieties, complete intersection Email: wmontoya@ime.unicamp.br\n\ntheorem ([7]) and the Hard Lefschetz theorem for projective orbifolds ([11]). When p = d + 1 − s the proof relies on the Cayley trick, a trick which associates to X a quasi-smooth hypersurface Y in a projective vector bundle, and the Cayley Proposition (4.3) which gives an isomorphism of some primitive cohomologies (4.2) of X and Y . The Cayley trick, following the philosophy of Mavlyutov in [7], reduces results known for quasi-smooth hypersurfaces to quasi-smooth intersection subvarieties. The idea in this paper goes the other way around, we translate some results for quasi-smooth intersection subvarieties to\n\nAcknowledgement. I thank Prof. Ugo Bruzzo and Tiago Fonseca for useful discus- sions. I also acknowledge support from FAPESP postdoctoral grant No. 2019/23499-7.\n\nLet M be a free abelian group of rank d , let N = Hom ( M, Z ) , and N R = N ⊗ Z R .\n\nif there exist k linearly independent primitive elements e\n\n, . . . , e k ∈ N such that σ = { µ\n\ne\n\n+ ⋯ + µ k e k } . • The generators e i are integral if for every i and any nonnegative rational number µ the product µe i is in N only if µ is an integer. • Given two rational simplicial cones σ , σ ′ one says that σ ′ is a face of σ ( σ ′ < σ ) if the set of integral generators of σ ′ is a subset of the set of integral generators of σ . • A finite set Σ = { σ\n\n, . . . , σ t } of rational simplicial cones is called a rational simplicial complete d -dimensional fan if:\n\nall faces of cones in Σ are in Σ ;\n\nif σ, σ ′ ∈ Σ then σ ∩ σ ′ < σ and σ ∩ σ ′ < σ ′ ;\n\nN R = σ\n\n∪ ⋅ ⋅ ⋅ ∪ σ t .\n\nA rational simplicial complete d -dimensional fan Σ defines a d -dimensional toric variety P d Σ having only orbifold singularities which we assume to be projective. Moreover, T ∶ = N ⊗ Z C ∗ ≃ ( C ∗ ) d is the torus action on P d Σ . We denote by Σ ( i ) the i -dimensional cones\n\nFor a cone σ ∈ Σ, ˆ σ is the set of 1-dimensional cone in Σ that are not contained in σ\n\nand x ˆ σ ∶ = ∏ ρ ∈ ˆ σ x ρ is the associated monomial in S .\n\nDefinition 2.2. The irrelevant ideal of P d Σ is the monomial ideal B Σ ∶ =< x ˆ σ ∣ σ ∈ Σ > and the zero locus Z ( Σ ) ∶ = V ( B Σ ) in the affine space A d ∶ = Spec ( S ) is the irrelevant locus.\n\nProposition 2.3 (Theorem 5.1.11 [5]) . The toric variety P d Σ is a categorical quotient A d ∖ Z ( Σ ) by the group Hom ( Cl ( Σ ) , C ∗ ) and the group action is induced by the Cl ( Σ ) - grading of S .\n\nNow we give a brief introduction to complex orbifolds and we mention the needed theorems for the next section. Namely: de Rham theorem and Dolbeault theorem for complex orbifolds.\n\nDefinition 2.4. A complex orbifold of complex dimension d is a singular complex space whose singularities are locally isomorphic to quotient singularities C d / G , for finite sub- groups G ⊂ Gl ( d, C ) .\n\nDefinition 2.5. A differential form on a complex orbifold Z is defined locally at z ∈ Z as a G -invariant differential form on C d where G ⊂ Gl ( d, C ) and Z is locally isomorphic to d\n\nRoughly speaking the local geometry of orbifolds reduces to local G -invariant geometry.\n\nWe have a complex of differential forms ( A ● ( Z ) , d ) and a double complex ( A ● , ● ( Z ) , ∂, ¯ ∂ ) of bigraded differential forms which define the de Rham and the Dolbeault cohomology groups (for a fixed p ∈ N ) respectively:\n\n(1,1)-Lefschetz theorem for projective toric orbifolds\n\nDefinition 3.1. A subvariety X ⊂ P d Σ is quasi-smooth if V ( I X ) ⊂ A #Σ ( 1 ) is smooth outside\n\nExample 3.2 . Quasi-smooth hypersurfaces or more generally quasi-smooth intersection sub-\n\nExample 3.2 . Quasi-smooth hypersurfaces or more generally quasi-smooth intersection sub- varieties are quasi-smooth subvarieties (see [2] or [7] for more details).\n\nRemark 3.3 . Quasi-smooth subvarieties are suborbifolds of P d Σ in the sense of Satake in [8]. Intuitively speaking they are subvarieties whose only singularities come from the ambient\n\nProof. From the exponential short exact sequence\n\nwe have a long exact sequence in cohomology\n\nH 1 (O ∗ X ) → H 2 ( X, Z ) → H 2 (O X ) ≃ H 0 , 2 ( X )\n\nwhere the last isomorphisms is due to Steenbrink in [9]. Now, it is enough to prove the commutativity of the next diagram\n\nwhere the last isomorphisms is due to Steenbrink in [9]. Now,\n\nH 2 ( X, Z ) / / H 2 ( X, O X ) ≃ Dolbeault H 2 ( X, C ) deRham ≃ H 2 dR ( X, C ) / / H 0 , 2 ¯ ∂ ( X )\n\nof the proof follows as the ( 1 , 1 ) -Lefschetz theorem in [6].\n\nRemark 3.5 . For k = 1 and P d Σ as the projective space, we recover the classical ( 1 , 1 ) - Lefschetz theorem.\n\nBy the Hard Lefschetz Theorem for projective orbifolds (see [11] for details) we\n\nBy the Hard Lefschetz Theorem for projective orbifolds (see [11] for details) we get an isomorphism of cohomologies :\n\ngiven by the Lefschetz morphism and since it is a morphism of Hodge structures, we have:\n\nH 1 , 1 ( X, Q ) ≃ H dim X − 1 , dim X − 1 ( X, Q )\n\nCorollary 3.6. If the dimension of X is 1 , 2 or 3 . The Hodge conjecture holds on X\n\nProof. If the dim C X = 1 the result is clear by the Hard Lefschetz theorem for projective orbifolds. The dimension 2 and 3 cases are covered by Theorem 3.5 and the Hard Lefschetz.\n\nCayley trick and Cayley proposition\n\nThe Cayley trick is a way to associate to a quasi-smooth intersection subvariety a quasi- smooth hypersurface. Let L 1 , . . . , L s be line bundles on P d Σ and let π ∶ P ( E ) → P d Σ be the projective space bundle associated to the vector bundle E = L 1 ⊕ ⋯ ⊕ L s . It is known that P ( E ) is a ( d + s − 1 ) -dimensional simplicial toric variety whose fan depends on the degrees of the line bundles and the fan Σ. Furthermore, if the Cox ring, without considering the grading, of P d Σ is C [ x 1 , . . . , x m ] then the Cox ring of P ( E ) is\n\nMoreover for X a quasi-smooth intersection subvariety cut off by f 1 , . . . , f s with deg ( f i ) = [ L i ] we relate the hypersurface Y cut off by F = y 1 f 1 + ⋅ ⋅ ⋅ + y s f s which turns out to be quasi-smooth. For more details see Section 2 in [7].\n\nWe will denote P ( E ) as P d + s − 1 Σ ,X to keep track of its relation with X and P d Σ .\n\nThe following is a key remark.\n\nRemark 4.1 . There is a morphism ι ∶ X → Y ⊂ P d + s − 1 Σ ,X . Moreover every point z ∶ = ( x, y ) ∈ Y with y ≠ 0 has a preimage. Hence for any subvariety W = V ( I W ) ⊂ X ⊂ P d Σ there exists W ′ ⊂ Y ⊂ P d + s − 1 Σ ,X such that π ( W ′ ) = W , i.e., W ′ = { z = ( x, y ) ∣ x ∈ W } .\n\nFor X ⊂ P d Σ a quasi-smooth intersection variety the morphism in cohomology induced by the inclusion i ∗ ∶ H d − s ( P d Σ , C ) → H d − s ( X, C ) is injective by Proposition 1.4 in [7].\n\nDefinition 4.2. The primitive cohomology of H d − s prim ( X ) is the quotient H d − s ( X, C )/ i ∗ ( H d − s ( P d Σ , C )) and H d − s prim ( X, Q ) with rational coefficients.\n\nH d − s ( P d Σ , C ) and H d − s ( X, C ) have pure Hodge structures, and the morphism i ∗ is com- patible with them, so that H d − s prim ( X ) gets a pure Hodge structure.\n\nThe next Proposition is the Cayley proposition.\n\nProposition 4.3. [Proposition 2.3 in [3] ] Let X = X 1 ∩⋅ ⋅ ⋅∩ X s be a quasi-smooth intersec- tion subvariety in P d Σ cut off by homogeneous polynomials f 1 . . . f s . Then for p ≠ d + s − 1 2 , d + s − 3 2\n\nRemark 4.5 . The above isomorphisms are also true with rational coefficients since H ● ( X, C ) = H ● ( X, Q ) ⊗ Q C . See the beginning of Section 7.1 in [10] for more details.\n\nTheorem 5.1. Let Y = { F = y 1 f 1 + ⋯ + y k f k = 0 } ⊂ P 2 k + 1 Σ ,X be the quasi-smooth hypersurface associated to the quasi-smooth intersection surface X = X f 1 ∩ ⋅ ⋅ ⋅ ∩ X f k ⊂ P k + 2 Σ . Then on Y the Hodge conjecture holds.\n\nthe Hodge conjecture holds.\n\nProof. If H k,k prim ( X, Q ) = 0 we are done. So let us assume H k,k prim ( X, Q ) ≠ 0. By the Cayley proposition H k,k prim ( Y, Q ) ≃ H 1 , 1 prim ( X, Q ) and by the ( 1 , 1 ) -Lefschetz theorem for projective\n\ntoric orbifolds there is a non-zero algebraic basis λ C 1 , . . . , λ C n with rational coefficients of H 1 , 1 prim ( X, Q ) , that is, there are n ∶ = h 1 , 1 prim ( X, Q ) algebraic curves C 1 , . . . , C n in X such that under the Poincar´e duality the class in homology [ C i ] goes to λ C i , [ C i ] ↦ λ C i . Recall that the Cox ring of P k + 2 is contained in the Cox ring of P 2 k + 1 Σ ,X without considering the grading. Considering the grading we have that if α ∈ Cl ( P k + 2 Σ ) then ( α, 0 ) ∈ Cl ( P 2 k + 1 Σ ,X ) . So the polynomials defining C i ⊂ P k + 2 Σ can be interpreted in P 2 k + 1 X, Σ but with different degree. Moreover, by Remark 4.1 each C i is contained in Y = { F = y 1 f 1 + ⋯ + y k f k = 0 } and\n\nfurthermore it has codimension k .\n\nClaim: { C i } ni = 1 is a basis of prim ( ) . It is enough to prove that λ C i is different from zero in H k,k prim ( Y, Q ) or equivalently that the cohomology classes { λ C i } ni = 1 do not come from the ambient space. By contradiction, let us assume that there exists a j and C ⊂ P 2 k + 1 Σ ,X such that λ C ∈ H k,k ( P 2 k + 1 Σ ,X , Q ) with i ∗ ( λ C ) = λ C j or in terms of homology there exists a ( k + 2 ) -dimensional algebraic subvariety V ⊂ P 2 k + 1 Σ ,X such that V ∩ Y = C j so they are equal as a homology class of P 2 k + 1 Σ ,X ,i.e., [ V ∩ Y ] = [ C j ] . It is easy to check that π ( V ) ∩ X = C j as a subvariety of P k + 2 Σ where π ∶ ( x, y ) ↦ x . Hence [ π ( V ) ∩ X ] = [ C j ] which is equivalent to say that λ C j comes from P k + 2 Σ which contradicts the choice of [ C j ] .\n\nRemark 5.2 . Into the proof of the previous theorem, the key fact was that on X the Hodge conjecture holds and we translate it to Y by contradiction. So, using an analogous argument we have:\n\nargument we have:\n\nProposition 5.3. Let Y = { F = y 1 f s +⋯+ y s f s = 0 } ⊂ P 2 k + 1 Σ ,X be the quasi-smooth hypersurface associated to a quasi-smooth intersection subvariety X = X f 1 ∩ ⋅ ⋅ ⋅ ∩ X f s ⊂ P d Σ such that d + s = 2 ( k + 1 ) . If the Hodge conjecture holds on X then it holds as well on Y .\n\nCorollary 5.4. If the dimension of Y is 2 s − 1 , 2 s or 2 s + 1 then the Hodge conjecture holds on Y .\n\nProof. By Proposition 5.3 and Corollary 3.6.\n\n[\n\n] Angella, D. Cohomologies of certain orbifolds. Journal of Geometry and Physics\n\n(\n\n),\n\n–\n\n[\n\n] Batyrev, V. V., and Cox, D. A. On the Hodge structure of projective hypersur- faces in toric varieties. Duke Mathematical Journal\n\n,\n\n(Aug\n\n). [\n\n] Bruzzo, U., and Montoya, W. On the Hodge conjecture for quasi-smooth in- tersections in toric varieties. S˜ao Paulo J. Math. Sci. Special Section: Geometry in Algebra and Algebra in Geometry (\n\n). [\n\n] Caramello Jr, F. C. Introduction to orbifolds. a\n\niv:\n\nv\n\n(\n\n). [\n\n] Cox, D., Little, J., and Schenck, H. Toric varieties, vol.\n\nAmerican Math- ematical Soc.,\n\n[\n\n] Griffiths, P., and Harris, J. Principles of Algebraic Geometry. John Wiley & Sons, Ltd,\n\n[\n\n] Mavlyutov, A. R. Cohomology of complete intersections in toric varieties. Pub- lished in Pacific J. of Math.\n\nNo.\n\n(\n\n),\n\n–\n\n[\n\n] Satake, I. On a Generalization of the Notion of Manifold. Proceedings of the National Academy of Sciences of the United States of America\n\n,\n\n(\n\n),\n\n–\n\n[\n\n] Steenbrink, J. H. M. Intersection form for quasi-homogeneous singularities. Com- positio Mathematica\n\n,\n\n(\n\n),\n\n–\n\n[\n\n] Voisin, C. Hodge Theory and Complex Algebraic Geometry I, vol.\n\nof Cambridge Studies in Advanced Mathematics . Cambridge University Press,\n\n[\n\n] Wang, Z. Z., and Zaffran, D. A remark on the Hard Lefschetz theorem for K¨ahler orbifolds. Proceedings of the American Mathematical Society\n\n,\n\n(Aug\n\n).\n\n[2] Batyrev, V. V., and Cox, D. A. On the Hodge structure of projective hypersur- faces in toric varieties. Duke Mathematical Journal 75, 2 (Aug 1994).\n\n[\n\n] Bruzzo, U., and Montoya, W. On the Hodge conjecture for quasi-smooth in- tersections in toric varieties. S˜ao Paulo J. Math. Sci. Special Section: Geometry in Algebra and Algebra in Geometry (\n\n).\n\n[3] Bruzzo, U., and Montoya, W. On the Hodge conjecture for quasi-smooth in- tersections in toric varieties. S˜ao Paulo J. Math. Sci. Special Section: Geometry in Algebra and Algebra in Geometry (2021).\n\nA. R. Cohomology of complete intersections in toric varieties. Pub-', lookup_str='', metadata={'source': '/var/folders/ph/hhm7_zyx4l13k3v8z02dwp1w0000gn/T/tmpgq0ckaja/online_file.pdf'}, lookup_index=0)]
使用 PyPDFium2
from langchain_community.document_loaders import PyPDFium2Loader
loader = PyPDFium2Loader("example_data/layout-parser-paper.pdf")
data = loader.load()
使用 PDFMiner
from langchain_community.document_loaders import PDFMinerLoader
loader = PDFMinerLoader("example_data/layout-parser-paper.pdf")
data = loader.load()
使用 PDFMiner 生成 HTML 文本
这对于将文本在语义上分块很有帮助,因为输出的 HTML 内容可以通过 BeautifulSoup
进行解析,以获取有关字体大小、页码、PDF 页眉/页脚等更结构化和丰富的信息。
from langchain_community.document_loaders import PDFMinerPDFasHTMLLoader
loader = PDFMinerPDFasHTMLLoader("example_data/layout-parser-paper.pdf")
data = loader.load()[0] # 将整个 PDF 作为单个文档加载
from bs4 import BeautifulSoup
soup = BeautifulSoup(data.page_content,'html.parser')
content = soup.find_all('div')
import re
cur_fs = None
cur_text = ''
snippets = [] # 首先收集所有具有相同字体大小的片段
for 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.text
snippets.append((cur_text,cur_fs))
# 注意:上述逻辑非常直接。人们还可以添加更多策略,例如删除重复片段(因为 PDF 中的页眉/页脚会出现在多个页面上,因此如果我们发现重复内容,则可以安全地假定它是多余信息)
from langchain.docstore.document import Document
cur_idx = -1
semantic_snippets = []
# 假设标题的字体大小高于其相应内容
for s in snippets:
# 如果当前片段的字体大小 > 前一个部分的标题 => 这是一个新标题
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(data.metadata)
semantic_snippets.append(Document(page_content='',metadata=metadata))
cur_idx += 1
continue
# 如果当前片段的字体大小 <= 前一个部分的内容 => 内容属于同一部分(如果需要,人们还可以创建子部分的树状结构,但这可能需要更多思考,可能是特定于数据的)
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
# 如果当前片段的字体大小 > 前一个部分的内容但小于前一个部分的标题,也创建一个新部分(例如,PDF 的标题将具有最大的字体大小,但我们不希望它包含所有部分)
metadata={'heading':s[0], 'content_font': 0, 'heading_font': s[1]}
metadata.update(data.metadata)
semantic_snippets.append(Document(page_content='',metadata=metadata))
cur_idx += 1
semantic_snippets[4]
Document(page_content='最近,为布局分析任务开发了各种深度学习模型和数据集。dhSegment [22] 利用完全卷积网络 [20] 进行历史文档的分割任务。基于目标检测的方法,如 Faster R-CNN [28] 和 Mask R-CNN [12] 用于识别文档元素 [38] 和检测表格 [30, 26]。最近,图神经网络 [29] 也被用于表格检测 [27]。然而,这些模型通常是单独实现的,没有统一的框架来加载和使用这些模型。\n近年来,人们对创建用于文档图像处理的开源工具表现出了极大的兴趣:在 Github 上搜索文档图像分析,可以找到 500 万个相关代码片段 6;然而,其中大多数依赖于传统的基于规则的方法或提供有限的功能。与我们的工作最接近的先前研究是 OCR-D 项目7,它也试图构建一个完整的 DIA 工具包。然而,类似于 Neudecker 等人开发的平台 [21],它是为分析历史文档而设计的,并不支持最新的深度学习模型。\nDocumentLayoutAnalysis 项目8 专注于通过分析存储的 PDF 数据处理出生数字 PDF 文档。像 DeepLayout9 和 Detectron2-PubLayNet10 这样的存储库是在布局分析数据集上训练的个别深度学习模型,没有支持完整的 DIA 流水线。Document Analysis and Exploitation (DAE) 平台 [15] 和 DeepDIVA 项目 [2] 旨在提高 DIA 方法(或 DL 模型)的可重现性,但它们没有得到积极的维护。OCR 引擎,如 Tesseract [14]、easyOCR11 和 paddleOCR12,通常不提供其他 DIA 任务(如布局分析)的全面功能。\n近年来,人们还做出了许多努力,创建了用于促进 DL 领域中可重现性和可重用性的库。像 Dectectron2 [35]、AllenNLP [8] 和 transformers [34] 为社区提供了完整的基于 DL 的支持,用于开发和部署通用计算机视觉和自然语言处理问题的模型。而 LayoutParser 则专门用于 DIA 任务。LayoutParser 还配备了一个社区平台,灵感来自已建立的模型中心,如 Torch Hub [23] 和 TensorFlow Hub [1]。它可以共享预训练模型以及专门用于 DIA 任务的完整文档处理流水线。\n已经有各种文档数据集以促进 DL 模型的开发。一些示例包括 PRImA [3](杂志布局)、PubLayNet [38](学术论文布局)、Table Bank [18](学术论文中的表格)、Newspaper Navigator 数据集 [16, 17](报纸图版布局)和 HJDataset [31](历史日文文档布局)。目前 LayoutParser 模型库中训练的一系列模型可支持不同用例。\n', metadata={'heading': '2 Related Work\n', 'content_font': 9, 'heading_font': 11, 'source': 'example_data/layout-parser-paper.pdf'})
使用 PyMuPDF
这是 PDF 解析选项中最快的,包含有关 PDF 及其页面的详细元数据,并返回每页一个文档。
from langchain_community.document_loaders import PyMuPDFLoader
loader = PyMuPDFLoader("example_data/layout-parser-paper.pdf")
data = loader.load()
data[0]
Document(page_content='LayoutParser: A Unified Toolkit for Deep\nLearning Based Document Image Analysis\nZejiang Shen1 (�), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain\nLee4, Jacob Carlson3, and Weining Li5\n1 Allen Institute for AI\nshannons@allenai.org\n2 Brown University\nruochen zhang@brown.edu\n3 Harvard University\n{melissadell,jacob carlson}@fas.harvard.edu\n4 University of Washington\nbcgl@cs.washington.edu\n5 University of Waterloo\nw422li@uwaterloo.ca\nAbstract. Recent advances in document image analysis (DIA) have been\nprimarily driven by the application of neural networks. Ideally, research\noutcomes could be easily deployed in production and extended for further\ninvestigation. However, various factors like loosely organized codebases\nand sophisticated model configurations complicate the easy reuse of im-\nportant innovations by a wide audience. Though there have been on-going\nefforts to improve reusability and simplify deep learning (DL) model\ndevelopment in disciplines like natural language processing and computer\nvision, none of them are optimized for challenges in the domain of DIA.\nThis represents a major gap in the existing toolkit, as DIA is central to\nacademic research across a wide range of disciplines in the social sciences\nand humanities. This paper introduces LayoutParser, an open-source\nlibrary for streamlining the usage of DL in DIA research and applica-\ntions. The core LayoutParser library comes with a set of simple and\nintuitive interfaces for applying and customizing DL models for layout de-\ntection, character recognition, and many other document processing tasks.\nTo promote extensibility, LayoutParser also incorporates a community\nplatform for sharing both pre-trained models and full document digiti-\nzation pipelines. We demonstrate that LayoutParser is helpful for both\nlightweight and large-scale digitization pipelines in real-word use cases.\nThe library is publicly available at https://layout-parser.github.io.\nKeywords: Document Image Analysis · Deep Learning · Layout Analysis\n· Character Recognition · Open Source library · Toolkit.\n1\nIntroduction\nDeep Learning(DL)-based approaches are the state-of-the-art for a wide range of\ndocument image analysis (DIA) tasks including document image classification [11,\narXiv:2103.15348v2 [cs.CV] 21 Jun 2021\n', lookup_str='', metadata={'file_path': 'example_data/layout-parser-paper.pdf', 'page_number': 1, 'total_pages': 16, 'format': 'PDF 1.5', 'title': '', 'author': '', 'subject': '', 'keywords': '', 'creator': 'LaTeX with hyperref', 'producer': 'pdfTeX-1.40.21', 'creationDate': 'D:20210622012710Z', 'modDate': 'D:20210622012710Z', 'trapped': '', 'encryption': None}, lookup_index=0)
使用 PyMuPDF
从目录加载 PDF
from langchain_community.document_loaders import PyPDFDirectoryLoader
loader = PyPDFDirectoryLoader("example_data/")
docs = loader.load()
使用 PDFPlumber
与 PyMuPDF 类似,输出的文档包含有关 PDF 及其页面的详细元数据,并且每页返回一个文档。
from langchain_community.document_loaders import PDFPlumberLoader
loader = PDFPlumberLoader("example_data/layout-parser-paper.pdf")
data = loader.load()
data[0]
Document(page_content='LayoutParser: A Unified Toolkit for Deep\nLearning Based Document Image Analysis\nZejiang Shen1 ((cid:0)), Ruochen Zhang2, Melissa Dell3, Benjamin Charles Germain\nLee4, Jacob Carlson3, and Weining Li5\n1 Allen Institute for AI\n1202 shannons@allenai.org\n2 Brown University\nruochen zhang@brown.edu\n3 Harvard University\nnuJ {melissadell,jacob carlson}@fas.harvard.edu\n4 University of Washington\nbcgl@cs.washington.edu\n12 5 University of Waterloo\nw422li@uwaterloo.ca\n]VC.sc[\nAbstract. Recentadvancesindocumentimageanalysis(DIA)havebeen\nprimarily driven by the application of neural networks. Ideally, research\noutcomescouldbeeasilydeployedinproductionandextendedforfurther\ninvestigation. However, various factors like loosely organized codebases\nand sophisticated model configurations complicate the easy reuse of im-\n2v84351.3012:viXra portantinnovationsbyawideaudience.Thoughtherehavebeenon-going\nefforts to improve reusability and simplify deep learning (DL) model\ndevelopmentindisciplineslikenaturallanguageprocessingandcomputer\nvision, none of them are optimized for challenges in the domain of DIA.\nThis represents a major gap in the existing toolkit, as DIA is central to\nacademicresearchacross awiderangeof disciplinesinthesocialsciences\nand humanities. This paper introduces LayoutParser, an open-source\nlibrary for streamlining the usage of DL in DIA research and applica-\ntions. The core LayoutParser library comes with a set of simple and\nintuitiveinterfacesforapplyingandcustomizingDLmodelsforlayoutde-\ntection,characterrecognition,andmanyotherdocumentprocessingtasks.\nTo promote extensibility, LayoutParser also incorporates a community\nplatform for sharing both pre-trained models and full document digiti-\nzation pipelines. We demonstrate that LayoutParser is helpful for both\nlightweight and large-scale digitization pipelines in real-word use cases.\nThe library is publicly available at https://layout-parser.github.io.\nKeywords: DocumentImageAnalysis·DeepLearning·LayoutAnalysis\n· Character Recognition · Open Source library · Toolkit.\n1 Introduction\nDeep Learning(DL)-based approaches are the state-of-the-art for a wide range of\ndocumentimageanalysis(DIA)tasksincludingdocumentimageclassification[11,', metadata={'source': 'example_data/layout-parser-paper.pdf', 'file_path': 'example_data/layout-parser-paper.pdf', 'page': 1, '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'})
使用 AmazonTextractPDFParser
AmazonTextractPDFLoader 调用 Amazon Textract 服务 将 PDF 转换为文档结构。目前,加载器仅执行纯 OCR,根据需求计划支持更多功能,如布局支持。支持单页和多页文档,最多支持 3000 页和 512 MB 大小。
为了成功调用,需要 AWS 账户,类似于 AWS CLI 的要求。
除了 AWS 配置外,它与其他 PDF 加载器非常相似,还支持 JPEG、PNG 和 TIFF 以及非原生 PDF 格式。
from langchain_community.document_loaders import AmazonTextractPDFLoader
loader = AmazonTextractPDFLoader("example_data/alejandro_rosalez_sample-small.jpeg")
documents = loader.load()
7. HTML
超文本标记语言(HyperText Markup Language 或 HTML) 是用于在网络浏览器中显示的文档的标准标记语言。
这里介绍了如何将 HTML
文档加载到我们可以在下游使用的文档格式中。
from langchain_community.document_loaders import UnstructuredHTMLLoader
loader = UnstructuredHTMLLoader("example_data/fake-content.html")
data = loader.load()
data
[Document(page_content='My First Heading\n\nMy first paragraph.', lookup_str='', metadata={'source': 'example_data/fake-content.html'}, lookup_index=0)]
使用 BeautifulSoup4 加载 HTML
我们还可以使用 BeautifulSoup4
来使用 BSHTMLLoader
加载 HTML 文档。这将从 HTML 中提取文本到 page_content
,将页面标题作为 title
存储在 metadata
中。
from langchain_community.document_loaders import BSHTMLLoader
loader = BSHTMLLoader("example_data/fake-content.html")
data = loader.load()
data
[Document(page_content='\n\nTest Title\n\n\nMy First Heading\nMy first paragraph.\n\n\n', metadata={'source': 'example_data/fake-content.html', 'title': 'Test Title'})]