Skip to main content
The Extraction Rules activity works with the recognized text of unstructured documents. The rules use named entities, keywords, and regular expressions to set conditions for the fields you need to extract.

​
Use cases

Add this activity to your document processing flow in the following cases:
  • If you have extracted several objects of the same type with the help of a Named Entities (NER) activity and now need to find a particular instance. For example, several Organization entities can be extracted from a lease agreement. You can use an Extraction Rules activity to determine the landlord and the tenant.
  • If you don’t have enough documents to use a Deep Learning activity. In this case, you can create an Extraction Rules activity and extract required objects by using keywords, regular expressions, or other business logic.
  • If you need to extract an object of a non-standard type. For example, you may want to extract a combination of date and time rather than date alone.
  • If you need to group several fields relating to the same object. For example, you can use an Extraction Rules activity to group together the name, date of birth, and address of each person. A Named Entities (NER) activity can also extract the names and addresses, but it won’t let you match addresses to their respective names.

​
How it works

This activity uses Natural Language Processing (NLP) to find search elements in the text. Search elements are named entities and words or phrases matching keywords or regular expressions. Together with fields extracted by other activities, search elements can then be used as building blocks for extraction rules, which specify the position of fields relative to other text. The extraction rules can be easily created via UI and further customized using a special language. Search elements can be mapped to fields, so that the recognized text from the region of a search element will be used to fill in its mapped field. For more information, see Setting up an Extraction Rules activity.