Learning from Few Samples: Lexical Substitution with Word Embeddings for Short Text Classification