October release

by Release notes

RELEASE NOTES

October release

We are pleased to present you the newest Metamaze features.

Do you have questions? You can always reach out to support@metamaze.eu. 

New features

  • Filters throughout the application will indicate through a number next to the filter icon how many filters are active
  • When opening a document for which entity extraction is still in progress, you will clearly see a progress indicator on the screen
  • When creating new users in Metamaze, automatic emails will be generated for the user to set their own passwords
  • Models will now be deprecated after 6 months when they haven’t been used to make predictions.
  • Jump from upload to upload when in the document view without the need to go back to the overview of uploads
  • Training speed for certain documents has been increased by ignoring pages without annotations

Our entire product team is very enthusiastic about these improvements. If you have any problems, questions or need advise, don’t hesitate to reach out to us via support@metamaze.eu.

CONTACT US

Questions or problems?

Don’t hesitate to reach out to support@metamaze.eu

Release 3.0.1

by Release notes

RELEASE NOTES

Release ML 3.0.1

We are pleased to present you Metamaze ML 3.0.1 In this release note, we will update you on all the major changes on the platform.

Do you have questions? You can always reach out to support@metamaze.eu. 

Model calibration

We have improved the way confidence scores (percentage scores you see in the UI for entities) are calibrated. Basically we made the scores more accurate.

What you see in the above images is that the left image shows the perfect model with the perfect data where the calculated scores perfectly match the accuracy (incase you were wondering, this does not exist in real life). The middle image shows our current solution to calculate the score. As you can see most of the scores ended up in the regions between 50% – 100%. The new calculations which are shown in the right image, show that the calculated scores are more in line with the accuracy.

What this means, for you as an admin, is that the accuracy scores in general will be lower than before. However this does not mean the model is doing worse. We tried to add a stronger correlation between scores and accuracy: lower scores are more likely to correspond to bad predictions.

As you are an admin in the Metamaze application, we recommend you to review the thresholds after your next training. It might be needed to lower them to obtain the same automation rate as before.

Training improvement

We have changed the way how models are trained. This new approach will further increase the training speed (see release 2.2 for more info on our previous improvements on this), how much it improves depends on the complexity and size of the datasets. An extra benefit with the new approach is that for smaller datasets the accuracy improves too!

Another improvement to increase the training speed is in the way we retried trainings in the past. When a training would get interrupted (network issues for example), the training would start from scratch for a training step. We now take advantage of the checkpoints which are created. When a training gets interrupted and needs to be retried, it will now start from the last known checkpoint instead of the start of a training step.

Our full product team is very enthusiastic about these improvements. If you have any problems, questions or need advise with the thresholds settings, don’t hesitate to reach out to us via support@metamaze.eu.

CONTACT US

Questions or problems?

Don’t hesitate to reach out to support@metamaze.eu

Release 2.2

by Release notes

RELEASE NOTES

Release 2.2

We are pleased to present you Metamaze 2.2. In this release note, we will update you on all the major changes on the platform.

Do you have questions? You can always reach out to support@metamaze.eu. 

1. New features in upload module

50+ languages support (BETA)

When putting documents through Metamaze, more than 50 languages are now supported for OCR, document classification, entity extraction, … . Below you can find a non-exhaustive list of supported languages:

Afrikaans, Albanian, Arabic, Armenian, Belarusian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Filipino, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Korean, Latvian, Lithuanian, Macedonian, Malay, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Russian, Serbian, Serbian, Slovak, Slovenian, Spanish, Swedish, Tamil, Thai, Turkish, Ukrainian, Vietnamese, Yiddish

with the following scripts:

Arabic, Armenian, Chinese, Cyrillic, Greek, Japan, Korean, Latin (including accents), Tamil, Thai

Extra filter option for queues

It is now possible to filter queues by the user who uploaded a batch.

2. Updates in integration

Sharing documents with Metamaze has been made easier by providing 2 new integration possibilities: SharePoint & SFTP.

These integrations are an on demand feature, when you need any of them, don’t hesitate to contact support@metamaze.eu

3. Annotation updates

Data masking

When privacy is of the utmost importance, a nifty feature has been added to Metamaze which allows you to mask sensitive data. The production data is handled as it should be, no special things happen there. When the document is sent towards the training set, however, the data is replaced with random but still relevant data. This is achieved by using the correct data masking type.

Merge occurrences

Sometimes you would like to merge all the occurrences of an entity into a single field for your output. From this release on this can be done by using the new setting in the entity configuration. Simply enable the option to get the merging going.

4. Training updates

Speed improvement

Our DevOps team has been working hard to improve the training process. They worked their magic and accomplished an astonishing 20 times speed improvement! Get those documents reviewed and train the model to get an improved model faster!

20x

speed improvement

5. Security updates

Okta login

We support Okta as an identity provider.

More than 14,000 global brands trust Okta to secure their digital interactions with employees and customers.

6. Various

  • The entity extraction and composite entity extraction have been improved to increase recognition.
  • Stability and scalability improvements.
  • Security improvements.

CONTACT US

Questions or problems?

Don’t hesitate to reach out to support@metamaze.eu

Release 2.1

by Release notes

RELEASE NOTES

Release 2.1

Task module revamped

The quality of your training data determines how well the predictions are of your model. It is, in other words, of utmost importance to perform regular quality checks especially before (re)training a model. We have created a – tasks – module around this where annotated documents can be easily reviewed. In this release, we have improved the ease of review by adding hints (“misannotation hints”). These will suggest annotations that might be missing or redundant. Also we have worked on simplifying annotation tasks by adding predictions to unlabelled documents once a training is present (“model assisted labelling”). This will help the user with annotations. By not annotating from scratch, the user can save a significant amount of (annotation) time.

  • – Redesign create task modal
  • – Model assisted labelling is added with annotation tasks (in case of a trained model)
  • – Misannotation hints are enabled with review tasks (in case of a trained model)

Example of a misannotation hint

Example when the hint was approved by the user (annotation updated)

Export to CSV

Besides JSON we support CSV as an output format for all data extracted, validated and enriched by Metamaze.

Other improvements

Use suggested tasks also when no training is present

We add suggested tasks when no training is present since before you train your first model, you will want to annotate uploaded documents so that afterwards you can review those annotated documents. What easier way is there than doing this using a task?

 

The review of a failed document is made more intuitive

When reviewing a failed document, you will be able to easily reuse the failed reason by confirming the failed status as well as unmarking the document as failed.

 

Release 2.0

by Release notes

RELEASE NOTES

Release 2.0

Creating your own custom document type, as a business user, has been a common practice in Metamaze. With this release, we go one step further and give the user the possibility to re-use an existing model via our interface. This is a powerful functionality for corporations who want to build further on work done for a specific department in a certain country or for partners who like to incorporate their own created model in client projects.

Document type is added as a new level

Document types are added to the overview page

Besides creating a new document type, it is also possible to add an existing document type to a project

For an existing document type, the user can choose per project which entities to reuse and/or new entities to add

Document type settings are introduced where organizations can be managed/added who have access to the doctype

Other improvements

The steps to create a project are simplified

Time needed to create a project is even lower than before

 

Name of logged-in user and organization are added to the navigation bar

Instead of ‘user’ and ‘system’ we now show the name of the user and the name of the organization

 

The number of documents in human validation and accuracy are added to the overview page

You no longer need to open every specific project to know the accuracy of the model(s) and/or amount of documents waiting to be validated

 

OCR speed has been increased by a factor 2x

Start annotating almost right after you have uploaded the training documents

 

New pipeline for training incremental models

The amount of time needed to train an incremental model has been drastically improved

 

Various performance improvements, security improvement and bug fixes

We are constantly working on improving our performance and security

 

Release 1.9

by Release notes

RELEASE NOTES

Release 1.9

Analytics module gets major overhaul

Having detailed analytics is crucial for having a full overview of what is automated and where there are improvement points. In this release, we made them more granular and way faster so you can track performance over time.

  • – Key metrics added
  • – Selection of a predefined timespan is made possible
  • – Details of status(es) are shown per doctype(s)
  • – Graphs are made more precise with indication of amount
  • – Last snapshot is added

Enrichments: a powerful pipeline step for integrating custom logic

Enrichments are a powerful new step in the pipeline that can be used for integrating any type of custom logic, custom code, external data lookups or even custom Machine Learning models.

  • – New step ‘enrichments’ is added to the pipeline, with full configuration through the project settings.
  • – Human validation module support for enrichments is added and updated live with updates
  • – Enrichments that are part of a composite entity are supported.
  • – Example enrichments are added to our public GitHub repos.

Other new functionalities

Automatic deletion of production data is made available

Automatic deletion of production data with configurable retention period ensures you can comply with GDPR and privacy regulation by not keeping data longer than is strictly needed.

Out-of-the-box support for e-mails

By supporting Outlook and IMAP integration out of the box, getting started with Metamaze for mailroom automation or incoming documents is easier than ever. So no more manual monitoring or custom RPA bot is needed to automatically process everything from orders@company.com!

Release 1.8

by Release notes

RELEASE NOTES

Release 1.8

"QA tasks" become "tasks"

  • Task types are introduced
  • Type ‘review’ has the same functionality as QA task
  • A new type ‘annotation’ is added, to allow for data control in the annotation process

Accelerate your annotations with suggested annotation tasks

  • Similar documents are grouped together
  • Documents are proposed that add the most value i.e. model learns from – MetaMaze Optimal Document Selection Strategy applied
  • Autocorrect allows detection of which documents might be misannotated

Get better training results based on your document type and language

  • choose between text or text and layout when creating a document type
  • choose between training one multi-lingual model or one model per language

Experience enhancements

  • Aggregated entity values are also visible in human validation
  • Tooltips are added to the project overview and action buttons
  • Automatically mark your batch as done in human validation and data annotation
  • Customization of view is added for tables in training and production

Release 1.7

by Release notes

RELEASE NOTES

Release 1.7

Focus on training

  • It’s now possible to send multiple relevant documents from production to training. 
  • Metamaze suggests QA tasks and ranks them by importance. 
  • Performance boost from additional training is measured. 

Better QA tasks

  • Button ‘mark as done’ and ‘failed’ have been added.
  • You are able to park a document for later/deeper review.
  • Clear view on missing, added and deleted labels during human validation.
  • You can filter by date.
  • Several UI/UX improvements.

Park & ride

  • When you’re in doubt on how a specific issue should be treated during human validation, you’ll be able to park it and add a comment for follow-up.
  • You can easily filter on parked documents.

Sticky filters

  • Filters are saved during human validation.
  • If no results have been found, a message is shown.
  • You see an indication when filtering is active.

Experience enhancements

  • Remove spaces and validation regex in project settings
  • Filter white pages out in the page management module.
  • The page management module shows which pages aren’t used.