Multi-Perspective Context Matching for SQuAD Dataset

Question answering is an important task in machine comprehension. The new SQuAD dataset allows us to deploy recent NLP deep learning techniques and train an end-to-end system to predict the start and end position of the answer in the given context, instead of precisely selecting the words of the correct answer. We propose to use combine bi-directional LSTM (BiLSTM) and context matching to develop a model for SQuAD dataset. We first use two BiLSTMs to encode the context and question word sequence. Then we apply context matching from multiple perspectives to produce a matching vector. Finally another BiLSTM is applied to the matching vector to predict the start and end positions. Several other tricks are also explored to enhance the prediction accuracy. Experimental results show that our model can achieve an F1 score of 61.27 and EM score of 49.50 on the development set.