Nyttiggörande avmaskininlärningsmodeller i verksamheten : Ökad metadatakvalitet med stöd från maskininlärning

Sammanfattning: Photographs, documents and other types of digitised data from the cultural heritage are collected in central databases to be made available to the public. These databases are known as aggregators. The aggregated data often have different purpose and formats, since they are created to suit the purpose of an individual institution. Metadata is data describing other data and is used to streamline the search through the different stored object within the aggregators.If all the stored metadata uses the same decided standard the search among the objects is quick and efficient. It is a common problem within aggregators that the stored metadata is of a lacking quality. When the quality of the metadata is lacking the search among the objects within the aggregator is slow, difficult and timeconsuming. The search may even give faulty results. In some cases data can go lost within large collections of data if the metadata is incorrect or missing. The knowledge about digitalisation and the resources to perform it, are often lacking in eg. a museum. This can sometimes lead to errors in the metadata. In 2019 a modell within machinelearning was developed during a project with the purpose to identify errors in the metadata of the swedish cultural heritage board’s aggregator K-samsök. In this study the modells ability to identify errors been evaluated. This evaluation was used to answer the following question: How can good quality of metadata be maintained whithin a organisation with support from a modell whithin machinelearning? This research contributes to the academy by informing the academy that there is still a problem that the quality of metadata in aggregators is of lacking quality. The research also provides suggestions for solutions to the problem, which in turn can give rise to further research. These solution suggestions are also of value to the Swedish National Heritage Board, as the study has been conducted with a focus on their aggregator K-samsök. The machine learning models can also be further developed and implemented by the Swedish National Heritage Board, which means that the models can provide value in the form of a basis to start from when improving the quality of the metadata stored in K-samsök