RetroSearch Browse

Home - News ( United States | United Kingdom | Italy | Germany ) - Football scores

Showing content from https://docs.unstructured.io/open-source/core-functionality/chunking below:

Chunking - Unstructured

Using Unstructured open source

Chunking functions in unstructured use metadata and document elements detected with partition functions to post-process elements into more useful “chunks” for uses cases such as retrieval-augmented generation (RAG).

Chunking BasicsChunking in unstructured differs from other chunking mechanisms you may be familiar with. Typical approaches start with the text extracted from the document and form chunks based on plain-text features, character sequences like "\n\n" or "\n" that might indicate a paragraph boundary or list-item boundary. Because unstructured uses specific knowledge about each document format to partition the document into semantic units (document elements), we only need to resort to text-splitting when a single element exceeds the desired maximum chunk size. Except in that case, all chunks contain one or more whole elements, preserving the coherence of semantic units established during partitioning. A few concepts about chunking are worth introducing before discussing the details.

Chunking is performed on document elements. It is a separate step performed after partitioning, on the elements produced by partitioning. (Although it can be combined with partitioning in a single step.)
In general, chunking combines consecutive elements to form chunks as large as possible without exceeding the maximum chunk size.
A single element that by itself exceeds the maximum chunk size is divided into two or more chunks using text-splitting.
Chunking produces a sequence of CompositeElement, Table, or TableChunk elements. Each “chunk” is an instance of one of these three types.

Chunking OptionsThe following options are available to tune chunking behaviors. These are keyword arguments that can be used in a partitioning or chunking function call. All these options have defaults and need only be specified when a non-default setting is required. Specific chunking strategies (such as by_title) may have additional options.

max_characters: int (default=500) - the hard maximum size for a chunk. No chunk will exceed this number of characters. A single element that by itself exceeds this size will be divided into two or more chunks using text-splitting.
new_after_n_chars: int (default=max_characters) - the “soft” maximum size for a chunk. A chunk that already exceeds this number of characters will not be extended, even if the next element would fit without exceeding the specified hard maximum. This can be used in conjunction with max_characters to set a “preferred” size, like “I prefer chunks of around 1000 characters, but I’d rather have a chunk of 1500 (max_characters) than resort to text-splitting”. This would be specified with (..., max_characters=1500, new_after_n_chars=1000).
overlap: int (default=0) - only when using text-splitting to break up an oversized chunk, include this number of characters from the end of the prior chunk as a prefix on the next. This can mitigate the effect of splitting the semantic unit represented by the oversized element at an arbitrary position based on text length.
overlap_all: bool (default=False) - also apply overlap between “normal” chunks, not just when text-splitting to break up an oversized element. Because normal chunks are formed from whole elements that each have a clean semantic boundary, this option may “pollute” normal chunks. You’ll need to decide based on your use-case whether this option is right for you.

ChunkingChunking can be performed as part of partitioning or as a separate step after partitioning: Specifying a chunking strategy while partitioningChunking can be performed as part of partitioning by specifying a value for the chunking_strategy argument. The current options are basic and by_title (described below).

from unstructured.partition.html import partition_html

chunks = partition_html(url=url, chunking_strategy="basic")

Calling a chunking functionChunking can also be performed separately from partitioning by calling a chunking function directly. This may be convenient, for example, when tuning chunking parameters. Chunking is typically faster than partitioning, especially when OCR or inference is used, so a faster feedback loop is possible by doing these separately:

from unstructured.chunking.basic import chunk_elements
from unstructured.partition.html import partition_html

url = "https://understandingwar.org/backgrounder/russian-offensive-campaign-assessment-august-27-2023-0"
elements = partition_html(url=url)
chunks = chunk_elements(elements)

# -- OR --

from unstructured.chunking.title import chunk_by_title

chunks = chunk_by_title(elements)

for chunk in chunks:
    print(chunk)
    print("\n\n" + "-"*80)
    input()

Chunking StrategiesThere are currently two chunking strategies, basic and by_title. The by_title strategy shares most behaviors with the basic strategy so we’ll describe the baseline strategy first: “basic” chunking strategy

The basic strategy combines sequential elements to maximally fill each chunk while respecting both the specified max_characters (hard-max) and new_after_n_chars (soft-max) option values.
A single element that by itself exceeds the hard-max is isolated (never combined with another element) and then divided into two or more chunks using text-splitting.
A Table element is always isolated and never combined with another element. A Table can be oversized, like any other text element, and in that case is divided into two or more TableChunk elements using text-splitting.
If specified, overlap is applied between chunks formed by splitting oversized elements and is also applied between other chunks when overlap_all is True.

”by_title” chunking strategyThe by_title chunking strategy preserves section boundaries and optionally page boundaries as well. “Preserving” here means that a single chunk will never contain text that occurred in two different sections. When a new section starts, the existing chunk is closed and a new one started, even if the next element would fit in the prior chunk. In addition to the behaviors of the basic strategy above, the by_title strategy has the following behaviors:

Detect section headings. A Title element is considered to start a new section. When a Title element is encountered, the prior chunk is closed and a new chunk started, even if the Title element would fit in the prior chunk.
Respect page boundaries. Page boundaries can optionally also be respected using the multipage_sections argument. This defaults to True meaning that a page break does not start a new chunk. Setting this to False will separate elements that occur on different pages into distinct chunks.
Combine small sections. In certain documents, partitioning may identify a list-item or other short paragraph as a Title element even though it does not serve as a section heading. This can produce chunks substantially smaller than desired. This behavior can be mitigated using the combine_text_under_n_chars argument. This defaults to the same value as max_characters such that sequential small sections are combined to maximally fill the chunking window. Setting this to 0 will disable section combining.

Recovering Chunk ElementsIn general, a chunk consolidates multiple document elements to maximally fill a chunk of the desired size. Information is naturally lost in this consolidation, for example which element a portion of the text came from and certain metadata like page-number and coordinates which cannot always be resolved to a single value. The original elements combined to make a chunk can be accessed using the .metadata.orig_elements field on the chunk:

>>> elements = [
...     Title("Lorem Ipsum"),
...     NarrativeText("Lorem ipsum dolor sit."),
... ]
>>> chunk = chunk_elements(elements)[0]
>>> print(chunk.text)
'Lorem Ipsum\n\nLorem ipsum dolor sit.'
>>> print(chunk.metadata.orig_elements)
[Title("Lorem Ipsum"), NarrativeText("Lorem ipsum dolor sit.")]

These elements will contain all their original metadata so can be used to access metadata that cannot reliably be consolidated, for example:

>>> {e.metadata.page_number for e in chunk.metadata.orig_elements}
{2, 3}

>>> [e.metadata.coordinates for e in chunk.metadata.orig_elements]
[<CoordinatesMetadata ...>, <CoordinatesMetadata ...>, ...]

>>> [
    e.metadata.image_path
    for e in chunk.metadata.orig_elements
    if e.metadata.image_path is not None
]
['/tmp/lorem.jpg', '/tmp/ipsum.png']

During serialization, .metadata.orig_elements is compressed into Base64 gzipped format. To deserialize .metadata.orig_elements, you can use the elements_from_base64_gzipped_json. For example:

from unstructured.partition.auto import partition
from unstructured.staging.base import elements_from_base64_gzipped_json

elements = partition('local-ingest-source/fake-email.eml', chunking_strategy='basic', include_orig_elements=True)

print("Before:\n")

for element in elements:
    metadata = element.metadata.to_dict()
    print(f"Element ID: {element.id}")
    print(f"  Compressed orig_elements: {metadata["orig_elements"]}")
    print("\n")

# Output:
# -------
# Before:
#
# Element ID: 083776ca703b1925e5fef69bb2635f1f
#   Compressed orig_elements: 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

print ("After:\n")

for element in elements:
    metadata = element.metadata.to_dict()
    print(f"Element ID: {element.id}")
    orig_elements = elements_from_base64_gzipped_json(metadata["orig_elements"])
    print(f"  Uncompressed orig_elements:")
    for orig_element in orig_elements:
        print(f"    {orig_element.category}: {orig_element.text}")
    print("\n")

# Output:
# -------
# After:
#
# Element ID: 083776ca703b1925e5fef69bb2635f1f
#   Uncompressed orig_elements:
#     NarrativeText: This is a test email to use for unit tests.
#     Title: Important points:
#     ListItem: Roses are red
#     ListItem: Violets are blue

Learn more Chunking for RAG: best practices

RetroSearch is an open source project built by @garambo | Open a GitHub Issue

Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo

HTML: 3.2 | Encoding: UTF-8 | Version: 0.7.4