A BERT-based Dual Embedding Model for Chinese Idiom Prediction - 42Papers