Learning Word Vectors Efficiently Using Shared Representations and Document Representations