If we focus the app for recommendations (movies, music, books, etc) with natural language inputs, the input may not contain complete information about the entity (e.g. it only mentions the name). We can query the knowledge graph and annotate the entities first to enhance the recommendation results.
In the UI we can ask the user to confirm which movie/book/music/other entity they meant if there's ambiguity; or auto-fill a template of metadata (e.g., movie name, director, genre, plot, production year, etc) and let user manually tweak the information if it looks wrong.