Texttract - Why does subsequent GetDocumentAnalysisAsync(getResultsRequest) Blocks have no relationships populated

Question

Hi and thank you for any help you can provide...

I am using Texttract with the .Net SDK.

I am able to successfully submit and get the results of a TABLE analysis job and now I am trying to loop through the results.  I can make my first call to GetDocumentAnalysisAsync then I loop through the blocks that are type CELL so I can get Row and Column ID and then I will look through all of the relationships cells to get the text.

This all works fine up to the point I need to get the next blocks with a call to GetDocumentAnalysisAsync passing the nexttoken ID.  I then get the next set of CELL block and begin looping again.  This time, none of my CELL blocks have any relationships populated to get the text from.  I just get what looks like a lot of empty cells.  I have verified the page is readable through the console demo page so there should be text there.

Here is my code (in c#) I am using to iterate through the blocks and retrieve the next set of blocks.

```

if (getResultsResponse.JobStatus == JobStatus.SUCCEEDED)
            {
                do
                {
                    getResultsResponse.Blocks.ForEach(x => {
                        if (x.BlockType.Equals("CELL"))
                        {
                            Console.WriteLine("Page: " + x.Page.ToString());
                            Console.WriteLine("Rowindex: " + x.RowIndex.ToString());
                            Console.WriteLine("Colindex: " + x.ColumnIndex.ToString());
                            x.Relationships.ForEach(y =>
                            {
                                y.Ids.ForEach(z =>
                                {
                                    var cellText = (from text in getResultsResponse.Blocks where text.Id == z.ToString() select text.Text).FirstOrDefault();
                                    if (!string.IsNullOrEmpty(cellText))
                                    {
                                        Console.Write($"{cellText} ");
                                    }
                                });
                            });
                        }
                    });

if (string.IsNullOrEmpty(getResultsResponse.NextToken)) { break; }

getResultsRequest.NextToken = getResultsResponse.NextToken;
                    getResultsResponse = await _textractClient.GetDocumentAnalysisAsync(getResultsRequest);
                } while (getResultsResponse.Blocks.Count > 0); 
}

```

Answer

To add to the above response, please review the below implementation for the `get_full_json` function in python. It implements the fix recommended.  
https://github.com/aws-samples/amazon-textract-textractor/blob/master/caller/textractcaller/t_call.py

Answer

Hi,

It seems according to your code that you are replacing the `getResultsResponse` in every loop by the subset gotten by the `NextToken`.
Usually the order of blocks in a full response in `WORD` then `LINE` then `CELL`, so at first iteration you have the WORD, LINE and CELL in the `getResultsResponse` which is why you can find the children elements. However when you get the next set of block, you will get only `CELL` so when you look at the children, they won't be in the subset.

ie, if the full response is `[ PAGE_1, WORD_1, WORD_2, TABLE_1, CELL_1, CELL_2]`, then the `getResultsResponse` will look like this if we accept 5 blocks by response before going to the next Token :
- 1st iteration : `getResultsResponse = [ PAGE_1, WORD_1, WORD_2, TABLE_1, CELL_1 ]` so you will find `WORD_1` the child of `CELL_1`
- 2nd iteration : `getResultsResponse = [ CELL_2 ]` so `WORD_2` child of `CELL_2` will not be part of this list, but is in the first iteration, which is why you cannot find it.

A fix for that would be to first query all the Response part, and concatenate in a single list. Then you can run you script to fetch the children within this concatenate list.

Hope this helps.

Answer

Collecting all the blocks up front in a Dictionary worked!  Thank you!

Here is the modified code...

```
            Dictionary blocks = new Dictionary();

if (getResultsResponse.JobStatus == JobStatus.SUCCEEDED)
            {
                do
                {
                    getResultsResponse.Blocks.ForEach(x => {
                        blocks.Add(x.Id, x);
                    });

if (string.IsNullOrEmpty(getResultsResponse.NextToken)) { break; }

getResultsRequest.NextToken = getResultsResponse.NextToken;
                    getResultsResponse = _textractClient.GetDocumentAnalysis(getResultsRequest);
                } while (getResultsResponse.Blocks.Count > 0);

foreach (KeyValuePair entry in blocks)
                {
                    if (entry.Value.BlockType.Equals("CELL"))
                    {
                        Console.WriteLine("Page: " + entry.Value.Page.ToString());
                        Console.WriteLine("Rowindex: " + entry.Value.RowIndex.ToString());
                        Console.WriteLine("Colindex: " + entry.Value.ColumnIndex.ToString());
                        entry.Value.Relationships.ForEach(y =>
                        {
                            y.Ids.ForEach(z =>
                            {
                                if (blocks.ContainsKey(z.ToString()))
                                {
                                    var cellText = blocks[z.ToString()].Text;
                                    if (!string.IsNullOrEmpty(cellText))
                                    {
                                        Console.Write($"{cellText} ");
                                    }
                                    else
                                    {
                                        Console.Write($"EMPTY ");
                                    }
                                }
                            });
                        });
                    }
                }

```

Texttract - Why does subsequent GetDocumentAnalysisAsync(getResultsRequest) Blocks have no relationships populated

相关内容