Research Article

PerAnSel:  A  Novel Deep Neural Network-Based System for Persian Question Answering

Figure 8

(1) The sentence is passed to a pretrained MBERT. (2) The [CLS] token (orange) is sent to a fully connected network. (3) Other tokens are sent to a LSTM network (green). (4) The output of LSTM and fully connected network are concatenated. (5) The final vector is returned as output.