The National Library’s collection contains Norwegian newspapers from the 1700s to the present day, and most of the newspaper material has been digitized. The digitized collection currently includes more than 4.6 million newspapers, in the form of scanned page images, XML files with textual content and information about the location of the text on the page, and raw text.
This digitized collection forms the basis for research in a range of disciplines, including history, social sciences, language, and language technology. When researchers work with the newspapers in the collection, they are typically interested in having the newspapers divided into individual articles. The current digitization pipeline includes a layout analysis that to some extent places text fragments from the same article in sequence, but the textual content of the newspapers is still “flat” (one newspaper is one text document). A varying number of columns per page, articles spanning multiple pages, different fonts, images, and advertisements all contribute to the complexity of the task.
We welcome a master's student who is interested in exploring news article segmentation; dividing a newspaper into the articles it consists of. The work could potentially increase the usefulness of the newspapers in the National Library’s collection.
The scope of the thesis can be somewhat tailored to the student’s own interests, but possible research topics include:
-
Evaluating existing multimodal models for article segmentation of Norwegian newspapers
-
Comparing text-based models with multimodal models, and possibly pure image segmentation models for article segmentation
-
Training and evaluating text-based models for article segmentation
-
Training and evaluating multimodal models for article segmentation
Since the material is in Norwegian, and we assume the student will need to do some manual data work (such as constructing an evaluation dataset), this project may be best suited for someone with Norwegian language skills. It will be possible to work from our offices at Solli in Oslo during parts of the thesis work.