A Local Attention-based Neural Networks for Abstractive Text Summarization.

Text summarization is a process of generating a concise summary of a given text. It is a popular research topic, and many studies have used sequence-to-sequence deep neural network models such as Long Short Term Memory (LSTM) or Gated Recurrent Unit (GRU) to tackle it. These models consist of two phases: encoding the input text and generating the summary. However, these models may lose information during the encoding phase, particularly when using deep layers, leading to inaccurate summaries. In this paper, we propose a bi-directional LSTM model with a Recurrent Residual Attention mechanism to address this issue. We tested our model on the Amazon Reviews dataset from the Stanford Network Analysis Project and found that it performed better than standard LSTM models and outperformed previous studies.