A development team is building an internal knowledge search tool. They want user queries and documents to be converted into numeric vector representations so that documents with similar meaning to a query can be retrieved, even when they do not share the exact same words. Which type of model should they use to generate these representations?