Document classification overview

What is document classification

Document Classification is a component in the Document Understanding^TM Framework that helps in identifying what types of files the robot is processing.

A file can be classified into one or more document types, depending on its content and the classification methods used:

• if a file contains a single logical document type (e.g., it is an Invoice or a Medical Record in all its entirety), then the classification component should be configured accordingly and return a single classification result;
• if a file contains multiple logical document types (e.g., it contains an Invoice from page 1 to page 5, and a Medical Record for the next 10 pages, and an Insurance Agreement from page 16 to the end), then the classification component should return multiple classification results, each corresponding to the right page range from the input file.

Document Types that classification is attempted on are the ones defined in the project Taxonomy.

When document classification should be used

On the one hand, if a project needs to process files that are all of the same document type and are always present as one instance per file (e.g., one invoice in one file), then classification is not necessary and can be skipped in its entirety.

On the other hand, if the project is dealing with two or more document types (e.g., the workflow must process Invoices and Medical Records which cannot be distinguished before processing), or files are sometimes expected to contain two or more distinct document types within them (e.g., one file contains 3 Invoices), then classification is strongly recommended.

How to use the document classification component

Classification is done through the Classify Document Scope activity. To classify the documents, you can use one or more classifiers, as the scope activity has the role of configuring and executing one or more algorithms for document classification and of offering an easy, unitary configuration option for all your needs.

In short, this is what the Classify Document Scope does:

• Provides all Classifiers (classification algorithms) the necessary configurations for them to run.
• Accepts one or more classifiers.
• Allows for document type filtering, taxonomy mapping, and minimum confidence threshold settings at classifier level.
• Reports classification information in a unified manner, irrespective of the source of classification.

The Classify Document Scope allows you to configure it by using the Configure Classifiers wizard. You can customize

• which document types are accepted from which classifier,
• what is the minimum confidence threshold for a given result that is acceptable for each classifier,
• what is the taxonomy mapping, at document type level, between the project taxonomy and the classifier's internal taxonomy (if any).

Please note that the order of the classifiers in the Classify Document Scope is important:

• classifiers are executed with priority, from left to right;
• a classification result returned by a classifier is accepted if it reports one of the acceptable document types and has a confidence threshold equal to or above the minimum confidence threshold set for that classifier;
• a classifier is executed with the page ranges that have remained unclassified by the previous classifiers only (so may be called multiple times in one execution).

Available classifiers

Based on the requirements of the use case, you can choose from several classification methods, called classifiers.

Classifiers can be found in the UiPath.IntelligentOCR.Activities or UiPath.DocumentUnderstanding.ML.Activities packages.

The available classifiers are:

• Keyword Based Classifier
• Intelligent Keyword Classifier
• Machine Learning Classifier
• Generative Classifier

You can always build your own Classifier, by using the public Document Processing Contracts, thus being able to implement any algorithm that fits your use case.