KB Labs

It's not just about data.
It's about connecting data with people.


KB Labs seeks to find new ways to combine the library’s digital cultural heritage collections and research, with the latest methods within machine learning. The lab is an initiative taken by the IT department at the Royal Danish Library.

Here you will find different applications made by the Royal Danish Library to visualize, engage or showcase the different materials or collections that we have available, to inspire and deepen the knowledge of what collections we actually have, and hopefully expand the use of these.

At the moment, applications and builds found here are considered experimental projects, and as such, can change or even be taken down without warning. This includes the data being presented by the applications.

Feel free to contact us, if you have any questions or wish to know more about a given project.


JUXTA Experimental project

Image collections presented as collages, with seamless zooming from full collection to full-screen single images. Context-sensitive meta data with link-back to the originating sources makes it easy to explore large collections.

Go to Juxta - Postcards
This collection visualizes the Royal Danish Library's collection: Postkortsamlingen (The Postcard Collection)

Go to Juxta - Maps
This collection visualizes the Royal Danish Library's collection: Kort og Atlas (The Maps and Atlas Collection)

DOTS Experimental project

DOTS visualizes Danish cities referenced in articles related to your search in the Royal Library's Danish newspaper archive.

Go to DOTS

PIXPLOT Experimental project

PixPlot is a product of Yale Digital Humanities Lab that uses Google Tensorflow to provide a spatial layout for a collections of images, where similar images are grouped together.

This works surprisingly well for all collections we have tried. In this demo, we used our collection of Dansk-Vestindien images. The clusters on the left hand side are automatically generated.

Extending PixPlot is on our to-do. Our primary needs are display of metadata for the individual images and rendering of the images in higher resolution.

Go to PixPlot

Adams Illustrated Panorama of History Experimental project

Adams Illustrated Panorama of History is a piece of educational art that is large measured physically (70 x 673 cm), electronically (1 gigapixel) and by information density (thousands of information fragments). Due to its size, viewing and talking about it is a challenge. We aim to handle that by using techniques such as OpenSeadragon for seamless overview-to-detail visualisation
  • Spatial bookmark support for referencing
  • Explorative meta-data for further reading (try hovering over the globe in the lower left corner)
  • Display-mode with random or selected areas of interest walks
The last two parts are under development, but this is labs so we publish early. Enjoy!

Go to Adams Illustrated Panorama of History

Netarchive Smurf Experimental project

Since 2004 The Danish Net Archive has been harvesting the Danish part of the internet multiple times every year.

For the period 1998 to 2004 we only have sparsely fragments of the Danish internet and data are not representative for these years.

The freetext search will search in html-pages for each year and the number of results found will be compared to the total number of html from that year. The total number of html pages in the archive is currently 20 billion (20,000,000,000).

Go to Netarchive Smurf

SMURF Experimental project

Smurf visualises how use of language in Danish newspapers has evolved since the 18th century.

Go to smurf

TAGS Experimental project

TAGS visualizes the use of HTML tags in the Royal Library's Danish Netarchive. In the app you can search and compare the use of different tags from 2007 until today.

Go to TAGS

FAKE NEWS PORTLET Experimental project
Machine Learning

This is a trained neural network, called a Deep Convolutional Generative Adverserial Network (DCGAN), which is comprised of a generator network and a discriminator network.

The generator has been trained to generate newspaper frontpages (for Aarhus Stiftstidende) as images, and the discriminator to tell whether a given image looks like a frontpage.

Go to Fake News Portlet

Lunch Experimental project

Images of lunch trays before and after lunch, with user controlled tempus vision: An experiment in visual exploration of paired image collections.

Go to Lunch

Official APIs at the Royal Danish Library Experimental project

Royal Danish libraries's open API for Aerial photograph, image and text collections.

Go to the APIs

SOUNDSEARCH Experimental project
Machine learning

Search in almost 100 danish TV and radio channels from 1990 to 2021 with a combined audio length over 100 years.

Upload an audio clip in Wav(.wav) or MP3(.mp3) format of duration less than 10 seconds and select the channel and year for the search. Search time can be up to a few minutes, we are currently working on reducing the search time :) Clicking results found matching the clip will be played in Mediestream and may require you need to login to gain access to that program.

The algorithm is designed to find matches in music but can in some cases also find speech also.

Go to Soundsearch

ZOOM Experimental project

A serendipitous presentation of 1 million newspaper pages from Mediestream. For your convenience, the 20 terapixels from the scanned papers are packed into a single image. Only thing needed is to zoom a bit.

Go to ZOOM

WORD2VEC Experimental project
Machine learning

Word2Vec is a high-dimensional word embedding based on an unsupervised machine learning algorithm using a simple neural network. It maps each unique word in a large text corpus to a vector.
The vector representation of the words reflects interesting semantic properties of the words. Words that appear in the same context will be close in the vector-space (similar words). But distance between words can also be used to find analogies. The word2vec demo features several corpora and a very large one based on over 65.000 Gutenberg E-books.

Go to Word2Vec

LOAR Experimental project

LOAR (Library Open Access Repository) is an open access repository for long term preservation of research data. LOAR also contains some of the Royal Danish Library’s open data sets.

Go to LOAR

MeLOAR Experimental project

MeLOAR is a dedicated front end for specific LOAR collections. MeLOAR displays a number of different collections, amongst them is "Beretningsarkiv for Arkæologiske Undersøgelser", which is an account archive for archaeological pdfs.

MeLOAR offers a keyword search and a location search, shows the search results with facets, maps if possible and highlights, and shows the highlights inside the pdfs as well.

Go to MeLOAR

Link graphs from the Danish Netarchive Experimental project

This archive contains harvested webpages from 1998 and up to today.
The link graphs show how domains are linking to another for several different queries.

Go to graph

Newspaper API Experimental project

Experimental API for publicly available data and metadata at the Royal Danish Library.

Currently the API delivers public data from the Royal Danish Library's newspaper collection. Data from the Danish newspaper has to be older than 140 years to qualify af "public data". The API is presented in the Swagger UI and can return data in JSON, JSONL and CSV. Requests to the API are based on search queries in the Mediestream-platform.

Go to the Newspaper API