Model improvements for the service 'Document Information Extraction' with architecture extensions such as pre-training or multi-language approaches