Intuitive and customizable AI solutions in the latest release

With the latest release, we have made significant progress to make the Konfuzio AI solutions even more customizable and intuitive. Customizable because we value the variety of business needs and preferences we are allowed to meet. Our products are designed to be intuitive to ensure you succeed effortlessly - without encountering unnecessary challenges.

New project setting - Select your preferred document viewer

Document Viewer Definition

A Document Viewer is the user interface for visualizing documents and their annotations. The Document Viewer is also used to interact with documents and prepare them for training. These interactions include manually adding annotations or providing feedback on AI-extracted annotations.

There are currently three different document viewers:

  • SmartView
  • TextView
  • DVUI

They generally fulfill the same purpose, but represent different generations of development and user-friendliness. The DVUI Document Viewer is the latest and most advanced. For each document, you can change the document viewer via the details page of the document. As we understand that each of our customers has individual preferences for a document viewer, with this release we are introducing a setting in the project that allows you to specify the default document viewer for all documents in a project. This means: All documents in the respective project are opened with the Document Viewer selected as part of this setting.

Project Settings - Document Viewer - edited Konfuzio Release

DVUI - Intuitive changing of labels for existing annotations

Providing human input by creating annotations or revising AI-extracted annotations is a central but sometimes cumbersome task for the successful training of AI models.

Therefore, we strive to improve the user experience and make the process more seamless by continuously conducting usability studies.

As a result of our latest findings, you can now change the label to which an annotation belongs with just a few clicks, without having to delete and reselect the annotation.

Example: Imagine that as part of your preparation for re-training your model, you are reviewing a document for which the extraction AI has already extracted annotations for the desired labels. During the review, you realize that one of the annotations has been incorrectly assigned to a label by the AI. To give the model useful feedback, you want to move the annotation to the correct label.

To do this, simply click on the annotation in the right-hand navigation area. A menu will then appear in the document and you can select the correct label set and the label itself. The short "how to" video illustrates the process:

DVUI - change label Konfuzio Release

New models for categorization AIs

Training an effective AI model is an art in itself. There are different base models with different architectures and sizes as well as several parameters that can be optimized. Determining the optimal combination of base model and parameter values is a non-trivial task that often requires a certain degree of trail-and-error.

To facilitate the creative process of selecting the right base models and parameters, Konfuzio makes it easy for you as a user to vary the basic model and the parameters. With the current release, we are introducing new basic models that can be used for categorization AIs.

These newly available BERT-based models significantly increase AI quality for large data sets.

However, please note that there is always a compromise: The increase in AI quality is accompanied by a longer processing time. This means that the new basic models are not necessarily the best choice. Always keep an eye on your data set and the project requirements.

-> You can find a complete list of changes in our server changelog

Review and outlook - look forward to further innovations

Helm & Nagel GmbH has made significant progress with Konfuzio in 2023.

A year of growth and innovation.

-> You can find a summary of 2023 and an outlook on what lies ahead in our blog post below

Questions, comments or criticism?

Write us a message:

    Maximilian Schneider Avatar

    Latest articles