Neural information extraction on technical short text: from theory to practical applications

Over 80% of all data in organisations is unstructured. A prevalent form of this data is technical short text, which captures vital information on events such as workplace accidents, traffic reports, and maintenance work orders. The short text contains valuable information, but existing pipeline-based information extraction (IE) systems are not directly usable by domain experts and rely on feature engineering. In this thesis we provide a solution to the challenge of domain agnostic IE on technical short text. We demonstrate how the process of IE can be reformulated from a domain-specific pipeline-based approach to a fully end-to-end deep learning-based process.