Algorithm Model

Composition

BiDAF Model

· Two bidirectional LSTM layers

· Self-attention mechanism

· Global attention mechanism

· Predict the start and end positions of the answer

R-Net Model

· Three bidirectional LSTM layers

· Bitwise attention mechanism

· Gated recurrent unit

· Predict the start and end positions of the answer

XLNet Model

· Neural network architecture based on Transformer

· Use autoregressive mechanism to predict answers

· Permutation-based language model