Use Machine Learning to Extract Information from Documents

0 %

Use Machine Learning to Extract Information from Documents

Details

Use Machine Learning to Extract Information from Documents

Juliana Morais

September 15, 2020

Created by

January 14, 2020

Beginner

15 min.

Get the machine learning model prediction for the documents you upload to Document Information Extraction.

You will learn

How to call and test Document Information Extraction
How to access and use Swagger UI (User Interface)
How to extract information from files with Document Information Extraction

The core functionality of Document Information Extraction is to automatically extract structured information from documents using machine learning. When you finish this tutorial, you will get field value predictions for the documents you upload to Document Information Extraction.

Step 1: Authorize Swagger UI

You will use Swagger UI, via any web browser, to call the Document Information Extraction APIs. Swagger UI allows developers to effortlessly interact and try out every single operation an API exposes for easy consumption. For more information, see Swagger UI.

In the service key you created for Document Information Extraction in the previous tutorial: Create Service Keys for Document Information Extraction, you should find (outside the uaa section of the service key) an entry called url and another entry called swagger (as highlighted in the image below).

To access the Document Information Extraction Swagger UI, add the swagger value (**/document-information-extraction/v1**) to the url value, paste it in any web browser and press Enter.
To be able to use the Swagger UI endpoints you need to authorize yourself. In the top right corner, click Authorize.
Get the access_token value created in the previous tutorial: Get OAuth Access Token for Document Information Extraction Using Any Web Browser, then add bearer in front of it, and enter in the Value field.
```
bearer <access_token>
```
Click Authorize, and then click Close.

Choose the correct value that needs to be entered when authorizing the Swagger UI for the Document Information Extraction API. "bearer access_token" "bearer clientid" "bearer tenantid" "bearer clientsecret"

Step 2: See list of document fields you can extract

Use the GET /capabilities endpoint to see a list of document fields you can extract with Document Information Extraction.

Click the endpoint name to expand it, click Try it out, and then Execute.

You should receive a response like the following:

If you get an error response code 401 (Unauthorized), your token is probably incorrect. Check if you have added the word bearer before the token and if the token value is complete and has been properly copied from the access_token value you received in the previous tutorial: Get OAuth Access Token for Document Information Extraction via Web Browser.

Done

Log on to answer question

Step 3: Create client

To use Document Information Extraction, you need to create a client. This client is used in most of the endpoints to distinguish and separate data.

You can either create a single client or multiple clients in the payload field of the POST /clients endpoint. The clientId values created here will be used in other service endpoints.

Expand the POST /clients endpoint.
Click Try it out.
Enter your clientId and clientName values in the payload field in the format you see in Examples for payload parameter (c_00 and client 00, for example).
Click Execute.

You should receive a response like the following:

CAUTION:

Be aware of the following Document Information Extraction trial account limitations:

Maximum 40 uploaded documents per week

Maximum 1 created clientId

Maximum 10 created enrichment dataIds

Done

Log on to answer question

Step 4: See created clients

To see a list of the clients you have created:

Expand the GET /clients endpoint.
Click Try it out.
Enter a maximum number of clients to be listed in the limit field.
Click Execute.

You should receive a response like the following:

Done

Log on to answer question

Step 5: Upload document

View Full Instructions

Document Information Extraction uses a globally pre-trained machine learning model that currently obtains better accuracy results with invoices and payment advices in English. The team is working to support additional document types and languages in the near future.

Upload to the service any document in PDF format that has content in headers and tables, such as an invoice, using the POST /document/jobs endpoint.

As an alternative to uploading your own document in PDF format to the service, you can use any of the following sample invoice files (right click on the link, then click Save link as to download the files locally):

Sample Invoice 1

Sample Invoice 2

Sample Invoice 3

Do the following:

Expand the POST /document/jobs endpoint.
Click Try it out.
Upload a document PDF file.

In options, enter the list of fields to be extracted from the uploaded file (documentNumber, taxId, purchaseOrderNumber, shippingAmount, netAmount, senderAddress, senderName, grossAmount, for example), the client id you created in step 3 (c_00, for example), and the document type (invoice, for example). In this case, you can use the following:

{
   "extraction":{
      "headerFields":[
         "documentNumber",
         "taxId",
         "purchaseOrderNumber",
         "shippingAmount",
         "netAmount",
         "senderAddress",
         "senderName",
         "grossAmount",
         "currencyCode",
         "receiverContact",
         "documentDate",
         "taxAmount",
         "taxRate",
         "receiverName",
         "receiverAddress"
      ],
      "lineItemFields":[
         "description",
         "netAmount",
         "quantity",
         "unitPrice",
         "materialNumber"
      ]
   },
   "clientId":"c_00",
   "documentType":"invoice",
   "receivedDate":"2020-02-17",
   "enrichment":{
      "sender":{
         "top":5,
         "type":"businessEntity",
         "subtype":"supplier"
      },
      "employee":{
         "type":"employee"
      }
   }
}

Click Execute.

After you have clicked Execute, you should receive a response like the following:

Copy the id from the Response body to see the result of the extraction in the next step.

Done

Log on to answer question

Step 6: See extracted fields

You can now use the GET /document/jobs/{id} endpoint to receive the prediction.

Expand the GET /document/jobs/{id} endpoint.
Click Try it out.
Set extractedValues to true to get the extracted values.
Enter the id received in the POST /document/jobs endpoint as the id.
Click Execute.

You should receive a response like the following:

In the response, you will find some general information about the document you uploaded. In headerFields, such as documentDate and taxAmount, and in lineItems, such as description and quantity, you will find the prediction for the extracted fields.

The prediction is made with a probability indicated by the confidence field which represents how certain the model is about its prediction. A confidence of 1 means that the model is 100% sure about its prediction.

If the status of the document (indicated by the status field) is PENDING instead of DONE, then it means that the service is still extracting some fields and the returned JSON file does not yet contain all the requested fields.

You have now successfully used our machine learning model to get field value predictions for the document you uploaded to Document Information Extraction.

Done

Log on to answer question

Step 7: Delete client

If you want to delete a client you created in Step 3, use the DELETE /clients endpoint.

Expand the DELETE /clients endpoint.
Click Try it out.
Enter in the payload field the client id you want to delete (c_00, for example).
Click Execute.