Nástroj pro strojové překládání komerčních dokumentů do různých jazyků

Abstract

This thesis addresses the design and implementation of a desktop application for automated translation of product XML feeds between language versions of e-shops. The theoretical part analyses the adoption of the Heureka XML standard on the Czech market, compares existing commercial tools (Mergado Translate, Conviu, LOCO) and evaluates translation API services DeepL and Google Translate. The practical part describes the application architecture, including use case, activity and class diagrams. The application is implemented in C# on the .NET Framework 4.7.2 platform with a WPF-based user interface. The solution includes batched translation via the DeepL API with sequential batch processing and throttling, optional product-level change detection (XML feed comparison via XmlReader), category mapping using CSV dictionaries, replacement of flags and availability tags, cleaning of platform-specific elements, externalised JSON configuration, and automated upload of the resulting feed to an SFTP server. The application’s functionality has been verified through long-term deployment in a production environment of an e-shop with approximately 7,500 products, 35,000 variants and 650 categories. Testing confirmed that the free DeepL API tier is sufficient for a medium-sized e-shop when change detection is used.