Input Text

• Text to extract entities from.
• Up to 50,000 characters long.

example: “The quick brown fox jumps over the lazy dog.”

Input Entities:

• An array of custom query entities to extract from the text, up to 30 entities per request.
• Each entity entry is described by a JSON with 3-4 key-value pairs:
  • “description”: a free text description of the entity, up to 150 characters long.
  • “type”: desired entity value output format, any primitive (“string”, “integer”, “float”, “boolean”), or any array of them (example: “array[string]”).
  • “var_name”: a descriptive entity variable name to be used in the output results, up to 50 characters long. It must start with a letter, followed by letters, digits, or underscores.
  • (optional) “valid_values”: an array of valid extracted entity values - use it to limit the extracted entity value to one of pre-defined possible values. Up to 20 values, up to 50 characters each.

example: [{“description: “number of animals mentioned in text”, “type”: “integer”, “var_name”: num_of_animals”}]

Output

• “results”: a JSON containing an entry for each input entity, mapping from var_name to to the extracted value.
• “stats”: a JSON with basic request statistics.

example: {“results”: {“num_of_animals”: 2}, “stats”: {“n_text_characters”: 44, “n_entities”: 1, “n_tokens_used”: 300}}

Features

• Extract custom entities from unstructured text.
• Powered by a powerful SOTA AI model.
• Multi-language support.
• Supports long texts: up to 50,000 characters.

Tips

• View our website for inspirational examples: https://www.textraction.ai/
• Input Text:
  • Remove any irrelevant parts of the text to focus the model on the relevant parts only (examples: HTML tags, irrelevant paragraphs, etc).
  • If relevant, add metadata and context for a better semantic understanding (example: “The following Curriculum Vitae was received from a candidate on 2023-04-23: …”).
• Input Entities:
  • Description:
    • Be explicit and accurately describe the desired value (example: “number of rooms in the property, including only bedrooms and living rooms”).
    • If relevant, specify an output format for better standartization (examples: YYYY-mm-dd, ISO, etc).
    • If needed, add limitations (example: “product summary, 3-5 words”).
  • Variable name:
    • Should be descriptive.
    • Think of them as variable names in a programming language, JSON keys, or columns names of a data table.
  • Type:
    • Should match the desired output value.
  • Valid values:
    • If needed, limit the extracted entity value to one of several expected values.
    • This is very useful when dealing with categorical values (example: automatically setting a value for a drop-down list or a radio button).
• Output:
  • The model is trained to handle missing/uncertain values by returning a “null” - handle them according to your product requirements (example: fill them with a default value).

Common Use Cases

• Parse any text:
  • Curriculum Vitae: candidate name, contact details, skills, education, etc.
  • Product listing: product name, specifications, price, etc.
  • Financial: revenues, number of sold items, earning per share, stock ticker, etc.
  • Customer support: order id, customer’s request, etc.
• Automatically fill/validate detailed user input fields (checkboxes, radio buttons, drop-down lists, text boxes, etc) based on a free text user input.
• Convert multiple texts into data tables:
  • Add filters based on text entities.
  • Train Machine Learning (ML) models over the extracted entities.
• Get answers to questions about a text in a structured format.

Pricing

We offer 100 requests per month for free to try our service. To use our API at a larger scale, subscribe to our PRO, ULTRA or MEGA plan at a price ranging from 0.017$ to 0.02$ per request. See exact details here.