Service: Cultural Heritage Tools - AI tools integrated in the National Library of Finland

Two AI tools have been integrated into the IT system of the National Library of Finland (digi.nationallibrary.fi): 1. Named entity recognition (NER) -Stanford NER software to recognize persons and place names from the Finnish historical newspapers. The first training and evaluation data set was created that consists of pages of both journals and newspapers from the time period 1836–1918. The Stanford NER system was trained using 220 annotated pages and evaluated using 34 pages. The results that were achieved were mostly good: 71 % of person names and 79 % of place names were recognized from the evaluation data. 2. Image classification -Google TensorFlow was used to classify images. A training dataset was created that was used to train the classifier model with TensorFlow. Using the trained model, it was possible to classify images into the following groups: advertisements, drawings, notes, pictures, paintings, maps or garbage. -The trained model was used to classify 50 000 images and its accuracy was about 70-75% when about 10 000 classified images were manually observed. -A graphical user interface was created that enabled an easier way to observe achieved results.