Machine Comprehension with MMLSTM and Clustering

摘要: We evaluate two new models for machine comprehension. MMLSTM builds on the work of Match-LSTM with Answer Pointer by adding more layers of Match-LSTMs that read both question and context one additional time. Question Clustering is based on the idea that there are many different types of questions, and that there may be a performance benefit based on clustering questions and training a different model for each cluster.