Reading Comprehension On SQuAD Using Tensorflow

This paper describes attempts at developing a high performance reading comprehension model. Reading comprehension, in the current context, is defined as extracting textual answers from context passages given qualitative text questions. To train and test our comprehension model, we use Stanford’s SQuAD dataset, a standard dataset for comprehension tasks. We present our initial implementation of the AttentiveReader model as first presented by Herman et al. We also introduce our use of Bi-Directional GRU cell’s as an extension to this model. We discuss the advantages of our chosen model in regard to the reading comprehension task. We also discuss some of its shortcomings. In order to overcome these shortcomings, we introduce and analyze several other models presented in various recently published research papers. Many of the discussed papers are currently ranked near the top of SQuAD’s leaderboard in regard to the ExactMatch and F1 scoring metrics.