Indexing Pipeline for Preprocessing

The indexing pipeline converts your files to documents, preprocesses them, and finally stores them in the DocumentStore. The query pipeline then uses the documents from the document store for search.

deepset Cloud offers preprocessing components that you can add to your indexing pipeline. This way, when the pipeline runs, your files are automatically converted, split, and cleaned. Pipeline templates available in deepset Cloud include indexing pipelines that preprocess TXT, PDF, MD, DOCX, PPTX, XLSX, XML, CSV, HTML, and JSON files out of the box. For other formats, you may need to preprocess outside of deepset Cloud.

Integrations for Processing Data

Unstructured

You can use unstructured.io through UnstructuredFileConverter to preprocess your files. To learn how to do this, see Use Unstructured to Process Documents.

Azure Document Intelligence

Use AzureOCRDocumentConverter to take advantage of Azure Document Intelligence preprocessing services. For details, see Use Azure Document Intelligence.

How to Prepare Your Files

Here's an outline of how to plan file preprocessing:

Your files determine which components to use in the indexing pipeline:

• If all your files are of one type, use a file converter appropriate for handling this type as the first component in your indexing pipeline. For supported converters, see Converters.
• If you have multiple file types, use FileTypeRouter as the first component in your indexing pipeline and connect it to the converters for all file types you need. FileTypeRouter classifies your files based on their extension and sends them to the converter that can best handle them.
  In this case, you'll also need DocumentJoiner to join the output of multiple components and send it as a single output to a preprocessor, such as DocumentSplitter.

The converter's task is to convert your files into documents. However, the documents you obtain this way may not be of the optimal length for the retriever you want to use and may still need cleaning up. PreProcessors are the components that handle the cleaning and splitting of documents. They remove headers and footers, which is useful for retaining the flow of sentences across pages; they delete empty lines and split your documents into smaller chunks.

Smaller documents speed up your pipeline. They're also optimal for vector retrievers, which often can't handle longer text passages. We recommend 100-word splits for vector retrievers. Keyword retrievers can work on slightly longer documents of around 200-300 words.

Use these suggestions as a starting point for your indexing pipeline. You may need to experiment with your settings to reach the optimal values for your use case.

Pipeline Components for Preprocessing

Components are very flexible, and you can use them in all types of pipelines, but several are typically used for indexing. Have a look at this table to help you choose the right components: