The Tailored-Media Project: Taming the data beast

The manual tagging of information such as persons, objects or places is a timeconsuming task that typically needs trained archivists. This inefficient use of resources leads to disproportionately high resource consumption for a relatively small quantity of tagged material. The TailoredMedia project set out to reverse this problem by using machine learning to tag a high quantity of material and having archivists ensure the quality of the tagged information. This paper describes the integration of a set of analysis tools based on artificial intelligence (AI), along with focused research on specific tools (efficient scene classification, fewshot object detection, scene text detection). The paper also introduces Taylor — an avatar for the system’s AI capabilities, designed as part of the user interface to provide explanations and support the user’s work to validate annotations and search for content.