In a Text Analytics service implementation, which natural language processing technique is best suited for identifying the primary topics within a set of documents?