Fig. 3From: Acceptance Prediction for Answers on Online Health-care CommunityArchitecture of our sentence encoder. The question and answer are first encoded by LSTM and no-linear layer as Q and A respectively, then Q and A are encoded by co-attention encoder [30] as CA. Next, context function maps CA into a vector h, i.e, the representation of the given pair of question and answer, which are fed into softmax layer for binary classification, i.e., accepted or notBack to article page