The environment variable eval.redo_ocr is relevant for Evaluation pipelines because it allows you to rerun OCR when running the pipeline to assess the impact of OCR on extraction accuracy. This assumes an OCR engine was configured when the ML Package was created. The other options are not valid environment variables for Evaluation pipelines.

References:  Document Understanding - Evaluation Pipelines

The Custom Named Entity Recognition out-of-the-box model is a machine learning package that allows you to bring your own dataset tagged with entities you want to extract from unstructured text. The model can be trained and deployed using the UiPath AI Center, and can be integrated with the UiPath Document Understanding framework. The model can be used to extract and classify text in various domains and formats, such as emails, letters, web pages, research papers, and call transcripts. For example, you can use the model to extract information such as names, dates, addresses, amounts, products, or any other custom entity from your documents. The model supports multiple languages and can be customized according to your needs.

References:  AI Center - Custom Named Entity Recognition ,  Custom Named Entity Recognition Documentation ,  UiPath AI Center - Document Understanding & Use of UiPath Custom Named Entity Recognition Model

To ensure the model is trained effectively, UiPath recommends that regular fields be labeled on at least 10 different pages in Data Manager before exporting the labeled documents to AI Center. This helps in providing enough variation and examples for the model to learn from and generalize effectively.

 Entity predictions refer to the process of identifying and highlighting a specific span of text within a communication that represents a value for a predefined entity type. For example, an entity type could be “Organization” and an entity value could be “UiPath”. Entity predictions are made by the platform based on the training data and the rules defined for each entity type.  Users can review, accept, reject, or modify the entity predictions using the Classification Station interface 1 2 .

References:  Communications Mining - Reviewing and applying entities ,  Communications Mining - Predictions - UiPath Documentation Portal .

According to the UiPath documentation, OmniPaqe is a paid OCR engine that requires a license to use. It is one of the most accurate and reliable OCR engines available, and it supports over 200 languages. The other OCR engines listed are free of charge, but they may have different features, limitations, and performance levels. For example, Tesseract is an open-source OCR engine that supports over 100 languages, but it may not be as accurate as OmniPaqe. Microsoft Azure OCR and Microsoft OCR are both cloud-based OCR engines that use Microsoft’s technology, but they have different capabilities and pricing models. Microsoft Azure OCR can process both printed and handwritten text, and it uses a pay-as-you-go model based on the number of transactions. Microsoft OCR can only process printed text, and it is included in the UiPath Studio license.

References:

Document Understanding - OCR Engines

Automation Pricing - Complete UiPath Enterprise Solution

Labels in UiPath Communications Mining are user-defined categories that can be applied to communications data, such as emails, chats, and calls, to identify the topics, intents, and sentiments within them 1 .  Labels are trained using supervised learning, which means that users need to provide examples of data that belong to each label, and the system will learn from these examples to make predictions for new data 2 . However, not all labels are equally easy to train, and some may require more examples than others to achieve good performance.  Labels that have bias warnings are those that have relatively low average precision, not enough training examples, or were labelled in a biased manner 3 . Precision is a measure of how accurate the predictions are for a given label, and it is calculated as the ratio of true positives (correct predictions) to the total number of predictions made for that label. A label with 100% precision means that all the predictions made for that label are correct, but it does not necessarily mean that the label is well-trained. It could be that the label has very few predictions, or that the predictions are only made on a subset of data that is similar to the training examples. This could lead to overfitting, which means that the label is too specific to the training data and does not generalize well to new or different data. Therefore, labels with 100% precision may still have bias warnings if they lack training examples, because this indicates that the label is not representative of the underlying data distribution, and may miss important variations or nuances that could affect the predictions. To improve the performance and reduce the bias of these labels, users need to provide more and diverse examples that cover the range of possible scenarios and expressions that the label should capture.

References :  1 :  Communications Mining Overview   2 : [Creating and Training Labels]  3 :  Understanding and Improving Model Performance  : [Precision and Recall] : [Overfitting and Underfitting] :  Fixing Labelling Bias With Communications Mining

The Data Extraction Scope activity provides a scope for extractor activities, enabling you to configure them according to the document types defined in your taxonomy. The output of the activity is stored in an ExtractionResult variable, containing all automatically extracted data, and can be used as input for the Export Extraction Results activity.  This activity also features a Configure Extractors wizard, which lets you specify exactly what fields from the document types defined in the taxonomy you want to extract 1 .

The extractors that can be used for Data Extraction Scope activity are:

Regex Based Extractor: This extractor enables you to use regular expressions to extract data from text documents.  You can define your own expressions or use the predefined ones from the Regex Based Extractor Configuration wizard 2 .

Form Extractor: This extractor enables you to extract data from semi-structured documents, such as invoices, receipts, or purchase orders, based on the position and relative distance of the fields.  You can define the templates for each document type using the Form Extractor Configuration wizard 3 .

Intelligent Form Extractor: This extractor enables you to extract data from semi-structured documents, such as invoices, receipts, or purchase orders, based on the labels and values of the fields. You can define the fields for each document type using the Intelligent Form Extractor Configuration wizard.

Machine Learning Extractor: This extractor enables you to extract data from any type of document, using a machine learning model that is trained on your data. You can use the predefined models from UiPath or your own custom models hosted on AI Center or other platforms. You can configure the fields and the model for each document type using the Machine Learning Extractor Configuration wizard.

References :  1 : Data Extraction Scope  2 : Regex Based Extractor  3 : Form Extractor : Intelligent Form Extractor : Machine Learning Extractor