อธิบายตัวโหลดเอกสาร

Large Language Models (LLMs) are powerful, but their knowledge is limited to their training data. To make AI agents truly useful, they often need access to fresh, external information.

• Imagine an agent that needs to answer questions about today's news.
• Or one that summarizes a specific document from your company's internal drive.

This is where data loading comes in.

Document Loaders are LangChain's way of bringing external data into your agent's workflow. They act as bridges, converting raw data from various sources into a standardized format that LLMs can understand.

Think of them as specialized data connectors. They handle the messy details of reading files, fetching web pages, or querying databases, and then present the data cleanly to LangChain.

When a loader fetches data, it wraps it into a Document object. This is LangChain's universal container for data.

Each Document typically has two main parts:

• page_content: The actual text content extracted from the source (e.g., the text from a PDF page, a web article).
• metadata: A dictionary containing extra information about the document (e.g., source URL, page number, file path, creation date).

One of the simplest ways to load data is from a plain text file on your computer. LangChain provides the TextLoader for this purpose.

It takes a file path, reads the content, and creates a Document object. Let's see how it works by creating a temporary file and loading it.

This example creates a temporary text file, writes some content, then uses TextLoader to read it and prints the result.

What if your data is on the internet? The WebBaseLoader is perfect for fetching content directly from web pages.

It's smart enough to extract the main text content, ignoring navigation, ads, and other irrelevant elements. You just provide the URL!

Here's how to fetch content from a simple example web page. Note: This loader typically requires beautifulsoup4 and lxml to be installed.

Loading data from PDFs is a common requirement. LangChain offers loaders like PyPDFLoader (which uses the pypdf library) to handle these.

These loaders are designed to extract text from complex formats, often preserving some structure. For PDFs, each page might become a separate Document object, with metadata indicating its page number.

There are also loaders for other formats like CSV, JSON, Markdown, and even specific data structures like Notion databases or Confluence pages.

Your data might live in databases. LangChain supports various database loaders to connect directly to your data sources:

• SQL Database Loader: Connects to relational databases (e.g., PostgreSQL, MySQL) to fetch data based on queries.
• MongoDB Loader: For NoSQL document databases.
• Elasticsearch Loader: To pull data from Elasticsearch indices.

These loaders allow your agents to dynamically query and retrieve data from your existing data infrastructure.

Which of the following is NOT a primary purpose of a LangChain Document Loader?

In this lesson, we explored LangChain's Document Loaders, essential tools for bringing external data into your AI agents.

• Loaders convert diverse data sources (text files, web pages, PDFs, databases) into LangChain's universal Document object.
• Each Document contains page_content and metadata.
• We saw practical examples with TextLoader for local files and WebBaseLoader for web content.

Next, we'll learn how to handle large documents by splitting them into manageable chunks.