Figure 3.

The structures of the WEM and CCM. WEM(CCM) obtains a new word (character)-level embedded feature vector after low-level feature complementation. The character (word)-level embedded feature vector obtains the feature weight through the activation function, and then the weight is multiplied with the character (word)-level embedded feature vector to extract the useful feature of the character (word)-level embedded. Finally, the new word (character)-level embedded feature vector is obtained through residual connection.