NLP table extraction tips
This Sensible Instruct method extracts a table in a document based on your description of the table title and each of its column headers.
Prompt Tips
-
Extract all columns to get the best results. If you describe only a few of the columns, your results may be less accurate.
-
Use the table titles or table column headers in the document as descriptions.
-
For more information about how to write descriptions, or "prompts", see Query Group.
-
For advanced options, see Advanced prompt configuration.
Examples
Example 1
The following example shows using the NLP Table method to extract data from a bank statement:
To try out this example in the Sensible app, take the following steps:
-
Navigate to the following example document:
-
Create fields to extract data using the following table:
Field name | Method | Overall table description | Column IDs and descriptions |
---|---|---|---|
savings_transaction_history | NLP Table | "savings transaction history" | date - "date" description - "description without totals" amount - "amount" |
Click the Send icon for each column.
- To verify the extracted data, scroll down in the right pane and compare the Extracted data section to the document in the left pane:
- (Optional) To standardize the representation of the extracted dates and dollar amounts, configure
date
andcurrency
types as shown in the following screenshots:
You should see that the formatting of the extracted data changes according to the types you specified. For example, Sensible reformats the date 04/11/23
to a standardized output format, 2023-04-11
:
Example 2
The following example shows using the NLP Table method to extract data from an auto insurance document:
To try out this example in the Sensible app, take the following steps:
-
Download the following example document:
Example document Download link -
Create a test document type in the Sensible app, then follow the prompts in the dialog to upload the example document and create the type.
-
Click Sensible Instruct and create prompts to extract data using the following table:
Field name | Method | Overall table description | Column IDs and descriptions |
---|---|---|---|
insured_vehicles_table | NLP Table | "insured vehicles" | manufacturer - "vehicle make (not model)" year - "year of manufacture" |
transactions_table | NLP Table | "transactions for insurance account" | transaction_date - "transaction date." transaction_description - "transaction description" |
For example, use the following screenshot as a guide for configuring the insured_vehicles_table
field:
Notes
For the full reference for this method in SenseML, see NLP Table.
Updated 17 days ago