Gated End-to-End Memory Network Based on Attention Mechanism

The memory network has been proved to work well on the problems of simple question answering tasks based on reasoning, such as factual reasoning. However, with the question answering tasks becoming more and more complex, for example, multi-fact question answering, and other dialog tasks, simple memory networks do not perform well on these problems because more interactions are required between memory and other modules. In this paper, a new memory network is proposed which combines the attention mechanism and the gated mechanism of the Gated End-to-End Memory Network, and achieves better results in complex question and answer tasks. Based on the End-to-End Memory Network, Gated End-to-End Memory Network uses the residual network and combines the output of the upper layer with the problem. This paper applies the attention mechanism to the Gated mechanism. The new method makes the distribution of attention in the memory network more accurate, which makes our experiment greatly be improved on the 20 bAbI-10k text question-answering dataset.