Toolbox

Classifiers

A classifier is a tool that routes documents to different template or model types. You establish classification criteria by defining keywords or providing training samples within the classifier. The classifier then uses these criteria to automatically match documents to your published custom templates or deployed information extraction models. After publishing the classifier, you can use a single classifier API to recognize and extract structured information from various types of business data. This saves you from manually classifying data before calling a specific template or model API. You can also adjust a classifier's categories by republishing it.

The classifier details page displays basic information, such as the classifier name, classifier ID, and number of categories. It also shows the training status and includes a Train button. Below this, the category management section contains a table listing the name, identifier, keywords, and number of training samples for each category. Each row has options to Edit Keywords, Edit Training Set, and Delete. You can add a new category by clicking the Add Category button.

Keyword: Unique text within the data of a specific category that is used to match a template or model with an input image. A keyword must appear only in the current template or model, which allows you to uniquely identify the category to which an image belongs. For example, if you create a classifier for ID cards and household registration books, you can select keywords for the household registration book, such as "Education Level", "Employer", or "Certificate Number". For the ID card, the only keyword that you can select is "Citizen Identity Number".

A training set improves classification accuracy for template-based categories. You can upload a ZIP file containing images that share the same layout. For a template with a complex layout, we recommend uploading at least 20 unique images of the same layout to form the training set.

Test: Once a classifier is trained, you can test its classification results.

Field types

A field type improves the accuracy and consistency of data extraction. You can define a field type by providing a dictionary of general, business, or industry-specific terms. During extraction, the system uses this dictionary to correct errors and perform format normalization, enhancing recognition accuracy and ensuring standardized output.

• General field: A system-predefined field type with common attributes.

• Custom field: A field type you create to meet your specific business needs. You typically define a custom field by providing a dictionary of enumerated values. Examples include lists of administrative divisions, candidate names, or days of the week.