Is there a node js code sample reflecting how to read multi line text within a table row using the Document AI form parser?

Question

I'd like to extract table data from a PDF file as done by the Try document AI demo. However, if a table row has multi line text within the same row, it's reflected as a separate row in my Json response. This is also reflected when using the Upload test document feature in the cloud console, as shown on the attached screenshot.Cloud Console demo.

However, the Try document AI demo (https://cloud.google.com/document-ai/docs/drag-and-drop) reflects the data correctly. Try document AI demo. Is this due to different API versions being used?

Node js code - using document AI v1beta3 version with form parser processor;

 var dataTable;
      for (const page of document.pages) {
        //tables
        for (const table of page.tables) {
         dataTable = printTableInfo(table, text);
         if (dataTable) return { table: dataTable };
        }
      }

// printTableInfo function
    function printTableInfo(table, text) {
     let dataArray = [];
     
       for (let i = 0; i < table.bodyRows.length; i++) {
      let bodyRowText = "";
      for (const bodyCell of table.bodyRows[i].cells) {
        const bodyCellText = getText(bodyCell.layout.textAnchor, text); //
        bodyRowText += `${JSON.stringify(bodyCellText.trim())} | `;
      
      }
      dataArray.push({
        [`Data`]: bodyRowText,
      });
      }
    
      return dataArray;
     
    }
// getText function
    const getText = (textAnchor, text) => {
      if (!text || !textAnchor) return;
   
      if (
        (!textAnchor && !textAnchor.textSegments && !textAnchor.textSegments) ||
        (textAnchor &&
          textAnchor.textSegments &&
          textAnchor.textSegments.length) === 0
      ) {
        return "";
      }
      const startIndex = textAnchor.textSegments[0].startIndex || 0;
      const endIndex = textAnchor.textSegments[0].endIndex;

      return text.substring(startIndex, endIndex);
    };

score 0 · Accepted Answer · answered May 02 '23 at 16:59

This could be related to which processor version you are using.

Form Parser has two Google-Managed versions:

pretrained-form-parser-v1.0-2020-09-23 (stable) [This is the default when creating a new processor]
pretrained-form-parser-v2.0-2022-11-10 (rc)

These versions may output different results due to changes in model quality.

The Try It Demo uses the rc version - You can see this on the "JSON" page in the app.

https://us-documentai.googleapis.com/v1/projects/224664002400/locations/us/processors/c2b2b49c83a94613/processorVersions/rc:process

Since you're getting the results as you expect from the rc processor version, I'd recommend changing the default processor version for your Form Parser, or changing the request code to send a request to that specific version.

Is there a node js code sample reflecting how to read multi line text within a table row using the Document AI form parser?

1 Answers1