Class TextExtraction

Namespace: LMKit.Extraction

Assembly: LM-Kit.NET.dll

Provides functionality to extract structured data from unstructured content using a language model.

public sealed class TextExtraction

Inheritance: object

TextExtraction

Inherited Members: object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Examples

Example: Extract customer information from text

using LMKit.Model;
using LMKit.Extraction;
using LMKit.Data;
using System;
using System.Collections.Generic;
LM model = LM.LoadFromModelID("llama-3.2-1b");
// Create the extractor
TextExtraction extractor = new TextExtraction(model);
// Define elements to extract
extractor.Elements = new List<TextExtractionElement>
{
new TextExtractionElement("CustomerName", ElementType.String, "Full name of the customer"),
new TextExtractionElement("Email", ElementType.String, "Customer email address"),
new TextExtractionElement("OrderTotal", ElementType.Double, "Total order amount"),
new TextExtractionElement("IsPremiumMember", ElementType.Bool, "Whether customer is a premium member")
};
// Set the content to extract from
extractor.SetContent(@"
Order Confirmation
Customer: John Smith (john.smith@email.com)
Premium Member: Yes
Total: $149.99
");
// Extract the data
TextExtractionResult result = extractor.Parse();
// Access results
Console.WriteLine($"Customer: {result["CustomerName"].Value}");
Console.WriteLine($"Email: {result["Email"].Value}");
Console.WriteLine($"Order Total: {result["OrderTotal"].Value}");
Console.WriteLine($"Premium: {result["IsPremiumMember"].Value}");
Console.WriteLine($"JSON: {result.Json}");

Example: Extract data from a PDF invoice

using LMKit.Model;
using LMKit.Extraction;
using LMKit.Data;
using System;
using System.Collections.Generic;
LM model = LM.LoadFromModelID("llama-3.2-1b");
TextExtraction extractor = new TextExtraction(model);
// Define invoice fields
extractor.Elements = new List<TextExtractionElement>
{
new TextExtractionElement("InvoiceNumber", ElementType.String),
new TextExtractionElement("InvoiceDate", ElementType.Date),
new TextExtractionElement("VendorName", ElementType.String),
new TextExtractionElement("TotalAmount", ElementType.Double),
new TextExtractionElement("LineItems", ElementType.StringArray, "List of items on the invoice")
};
// Load PDF and extract
var pdfAttachment = new Attachment("invoice.pdf");
extractor.SetContent(pdfAttachment);
TextExtractionResult result = extractor.Parse();
Console.WriteLine($"Invoice #{result["InvoiceNumber"].Value} from {result["VendorName"].Value}");

Example: Extract with enum values

using LMKit.Model;
using LMKit.Extraction;
using LMKit.Data;
using System;
using System.Collections.Generic;
LM model = LM.LoadFromModelID("llama-3.2-1b");
TextExtraction extractor = new TextExtraction(model);
// Define element with allowed values
var sentimentElement = new TextExtractionElement("Sentiment", ElementType.String, "Customer sentiment");
sentimentElement.Format.AllowedValues = new List<string> { "Positive", "Neutral", "Negative" };
extractor.Elements = new List<TextExtractionElement>
{
new TextExtractionElement("CustomerFeedback", ElementType.String),
sentimentElement
};
extractor.SetContent("The product exceeded my expectations! Great quality and fast shipping.");
var result = extractor.Parse();
Console.WriteLine($"Sentiment: {result["Sentiment"].Value}"); // Output: Positive

Remarks

The TextExtraction class enables you to define a set of elements to extract from various content sources, including plain text, images, PDF documents, HTML files, Microsoft Office formats (DOCX, XLSX, PPTX), and email files (EML, MSG, MBOX). It leverages a language model to parse the content and extract the specified elements into structured data.

For a complete list of supported file formats, see the Attachment class documentation.

Key Features

Define extraction elements with names, types, and descriptions
Extract from text, images, PDFs, and Office documents
Support for various data types (string, integer, decimal, boolean, date, enum, arrays)
Optional OCR engine integration for scanned documents
Guidance text to improve extraction accuracy
JSON output with structured results

Typical Workflow

Create a TextExtraction instance with a language model
Define elements to extract using Elements
Set the content source using SetContent(string) or similar methods
Call Parse(CancellationToken) to extract the data
Access results via TextExtractionResult

Constructors

TextExtraction(LM): Initializes a new instance of the TextExtraction class with the specified language model.

Fields

Description: Gets or sets the description for the current extraction schema. This value is automatically populated from the schema's "description" field when calling SetElementsFromJsonSchema(string).

Title: Gets or sets the title for the current extraction schema. This value is automatically populated from the schema's "title" field when calling SetElementsFromJsonSchema(string).

Properties

Elements: Gets or sets the list of TextExtractionElement instances that define which elements to extract from the content.

Guidance: Gets or sets semantic guidance for the extraction process.

HumanVerificationThreshold: Gets or sets the confidence threshold below which HumanVerificationRequired is set to true.

JsonSchema: Gets the JSON schema representation of the extraction elements.

MaximumContextLength: Gets or sets the maximum context length (in tokens) allowed for the language model during text extraction.

Model: Gets the language model instance used to drive the extraction process.

NullOnDoubt: Gets or sets a value indicating whether the language model should return null for uncertain content rather than risk an aggressive extraction that could lead to false positives.

OcrEngine: Gets or sets an optional OcrEngine used to perform traditional OCR on raster content.

OverflowStrategy: Gets or sets the strategy applied when the content does not fit the context window a single inference pass can hold.

PreferredInferenceModality: Gets or sets the preferred modality for inference. This determines whether text, image, or both modalities are used when processing input. Defaults to Multimodal.

StopWhenAllFieldsFound: Gets or sets whether a windowed extraction stops as soon as every field the schema can complete has a value, instead of reading every remaining window.

Methods

ClearContent(): Removes all previously set input (both text and attachments) so that no content remains for extraction.

Parse(CancellationToken): Parses the content synchronously to extract the defined elements.

ParseAsync(CancellationToken): Parses the content asynchronously to extract the defined elements.

SchemaDiscovery(CancellationToken): Discovers an optimal JSON Schema for the current input content.

SchemaDiscoveryAsync(CancellationToken): Asynchronously discovers an optimal JSON Schema for the current input content.

SetContent(Attachment): Adds all pages of an attachment to be processed for data extraction.

SetContent(Attachment, int): Adds a specific page of an attachment to be processed for data extraction.

SetContent(Attachment, string): Adds specified pages of an attachment to be processed for data extraction.

SetContent(ImageBuffer): Sets the content for extraction from the specified image buffer.

SetContent(IEnumerable<Attachment>): Adds multiple attachments to be processed for data extraction.

SetContent(string): Sets the text content from which elements will be extracted.

SetElementsFromJsonSchema(string): Configures the text extraction elements by parsing a JSON schema.

Events

Progress: Occurs when the extraction operation makes progress, including phase transitions and pass completions.

Table of Contents