Data quality for Numismatics based on Natural language processing and Neural Networks - D4N4

点击翻译按钮获取中文摘要

The aim is to implement research tools for numismatics that are also suitable for other subjects that make use of images and/or writing and are available in large quantities. This involves methods of a) natural language processing (NLP) for multilingual and non-uniform coin descriptions and their combination with a hierarchical iconographic thesaurus, as well as b) image recognition of individual elements and overall compositions using deep learning, a component of artificial intelligence (AI). Both approaches have already been prototypically developed and tested and have already delivered helpful results, especially in increasing data quality. However, before deployment and subsequent use by others, the tools developed must be expanded and improved upon. This will require an increase in the amount of data, which is to be achieved by indexing archive holdings and recording the material available on the Internet. The implementation of the tools together with the enrichment of the data will constitute the creation of a comprehensive corpus of ancient coinage of Thrace and subsequently allow for its monetary and iconographic analysis.The project has a pilot character and is integrated into the international efforts of a type repository of ancient Greek coins according to historical landscapes. The circle of users for the tools is correspondingly large in the numismatic community and, beyond that, for other researchers of antiquity working on vases, reliefs, gems, etc. Furthermore, the implementation of quality controls, including the above-mentioned procedures, is also a matter of ensuring the scientific quality of the large amounts of data, from which other projects can also benefit.By providing the data sets used, as well as our results in the field of Deep Learning, we also offer the opportunity for other computer scientists to test out the data/results as a benchmark for new approaches.