Towards incremental learning of word embeddings using context informativeness