Query in Uipath Document Understanding

Help Something Else
1

Hi Experts , Since i’m new to DU . I got a situation to process 1000 different formats of documents using DU. Is it possible to create Du BOT for this case? kindly suggest me

2

Yes, it is possible to create a DU (Document Understanding) bot to process 1000 different formats of documents. Document Understanding is a powerful feature in UiPath that allows you to extract structured data from unstructured documents such as invoices, receipts, contracts, etc.

To handle multiple document formats, you would typically follow these steps:

  1. Document Understanding Setup: Set up Document Understanding in your UiPath environment. This involves creating document types and training the machine learning models to recognize and extract data from specific document layouts.

  2. Document Classification: Train a document classifier model to identify the document type or format. This model will determine which extraction model to use for a particular document.

  3. Extraction Models: Create extraction models for each document format. An extraction model is trained to identify and extract specific data fields (such as customer name, invoice number, date, etc.) from a document. You will need to create an extraction model for each of the 1000 document formats you mentioned.

  4. Training Data: Gather a diverse set of documents for each format and create training data by manually labeling the data fields you want to extract. This labeled data will be used to train the extraction models.

  5. Train Extraction Models: Use the labeled data to train the extraction models for each document format. This involves configuring the extraction scope, defining the data fields, and training the models using the Document Understanding ML Trainer.

  6. Integration with RPA Workflow: Integrate the DU bot into your RPA workflow to process the documents. This typically involves using the DU activities in UiPath Studio, such as the “Classify Document” activity to classify the document type, and the “Process Document” activity to extract data using the appropriate extraction model.

  7. Validation and Iteration: Test the bot with sample documents of different formats and review the extracted data. Make adjustments to the extraction models and refine the training data as needed to improve the accuracy of the extraction.

It’s important to note that handling 1000 different document formats can be a complex and time-consuming task. It may require significant effort to create and train the extraction models for each format. You might consider prioritizing the most frequently encountered document formats or grouping similar formats together to reduce the number of extraction models.

UiPath provides extensive documentation and resources on Document Understanding, including step-by-step tutorials and example workflows, which can assist you in implementing this solution.

If you’re new to UiPath and Document Understanding, it might be helpful to seek guidance from UiPath experts or consider engaging with UiPath’s professional services or partners who have experience with complex Document Understanding projects.

Good luck with your document processing project!

3

Hi @Vajrang thanks for your response. if possible can we set up a call related to this to get more idea from you for handling the scenario in your free time only and if you wish to have a connect or else i don’t want you to disturb you.

Regards

4

Hey,

Yes, it is possible to solve these kinds of situations through Document Understanding, you can create a bot to process 1000 different formats of documents.

Follow Document Understanding Framework steps:

  1. Taxonomy - Define Document type and fields for extraction.

  2. Digitize - if your documents are scanned images or raster PDFs, you will need OCR (Optical Character Recognition) technology to transform the image into text.

  3. Calssify - Classifying Documents(It is “Invoice” “Receipt” / “Resume”)
    Are there multiple document types in your taxonomy? The robot must identify which document category the current case falls into and process it accordingly.

  4. Validate - It does not provide you a 100% accuracy. In such cases, it also shows the percentage of accuracy level for the classification provided. If this value is below a target, a human can be asked for confirmation. The process will resume after the human has confirmed/approved the classification.

  5. Train - They require initial training and can also be further trained to improve their accuracy. can be used as a retraining dataset that in time will have better accuracy, requiring less human intervention.

  6. Extract - There are two extraction method
    1) Rule Based
    2) ML Based

  7. Export - Now you have the exact data, as you defined it in your taxonomy, you know which document type you’re processing, what the extracted entities (e.g. invoice ) and the corresponding fields (e.g. invoice number, total, Billed To, etc.).

Hope it helps.
Regards

1 Like
5

Hi @kajal_ujjainwal , thanks for your response,
In my case there are totally 10+ categories like tax, refund, property tax, overpayments ,amount due forms, Adjustment forms etc., not the general ones like receipt, invoices, bill
So under there are 10 different categories no standard formats for one category. All documents have different format not unique ones.

So in realtime bot needs to process 40,000 docs mostly all are different formats.

So in this scenario how we can give keywords to classify 1000+ formats and if i try give keywords for some formats the same keywords also present in another docs format. Classification not done properly or wrongly.And extraction will be complex one while developement

Kindly suggest me any solution.

6

@nithish.dhanabal1

When you say 1000 different formats…is there no similarity between any of them?

We consider different when all most of the document looks completely different but not when data is different

Is that same?

If all are different then this model shpuld be trained with as many as possible and then have human intervention in the bot to add more evry now and then when a complete nee format comes in

Cheers

7

@Anil_G yes not even keywords are similar for all docs under one category so only classification not done properly.

That’s how i got the docs from the clients. But fron there side even though it is not similar format but it comes under one category.

Its like under one category 10+ different formats not even similar keywords for all docs under one category.

8

Hey,

Try with multiple keywords.

9

@Anil_G and if we classify somewhere using most similar keywords but at the extraction part will be complex if we not get atleast no of similar formats.
Most of the docs remains unable to extract similar data right?

Pls suggest if you have any solutions to handle 1000+ formats

As per my knowledge keywords should be unique and available in all docs under one category but this is not that case.
Is my understanding right?

10

@kajal_ujjainwal just because the formats are different under one category. The keyword sometime not found in docs and wrongly classified as different doc type due to that keyword is also present in another doc type.

The tough part is we can’t identify unique keywords because we are facing multiple formats under one doc type

11

@nithish.dhanabal1

Keywords need not be similar…we can go with different as well…like it would classify as different docs but those classifications can be considered as one type for the use case…like say keyword 1 and 2 are type 1 and keyword 3 and 4 are type 2…both types can be taken as one vendor and that is not a problem it is the matter of your logic that you implement…

as part of formats if formats are 10+ then you need to train all 10+ with multiple files of each atleast 5 samples for each type

if you say very format of all 1000 are different then it needs to be a continuous improvement model with proper human intervention as fr every new frmat there needs to be a training or validation

cheers

cheers