Language agnostic architecture for a search engine

A computer-implemented method, comprising:

• receiving, in an online marketplace, from a client device used by a media producer, a media file, wherein the online marketplace stores multiple media files;
• receiving, from the media producer, a metadata associated with the media file, the metadata including a textual description of a media file content;
• associating a one or more words in the textual description of the media file content with an unambiguous concept according to a knowledge database, wherein the knowledge database is a multilingual resource;
• storing, in a database, a map linking the media file with the unambiguous concept;
• forming a data structure in a target language from the map by identifying a parent label and a child label for each media file, wherein the parent label has a larger number of annotations in the target language relative to the child label, and wherein a link joining a parent node associated with the parent label to a child node associated with the child label indicates a navigation route for a search engine to move from the parent node to the child node in search of the media file;
• and
• executing, by the search engine, a traversal of the data structure to retrieve the media file in response to a query made in the target language, wherein the traversal is constrained to parent-to-child links for which the parent node has a greater number of target-language annotations than the child node.