Research On Chinese Automatic Text Summarization Based On Sea2Seq Model

In the information age with rapid development of Internet technology,people have to face the explosive growth of Internet information on the network day by day.Due to the quick pace of life,people usually read the summary before they read an article so that they can comprehend the main idea and information of th earticle quickly.Summary is a concisely natural language paragraph which can reflect the articles' ideas and contents coherently and accurately.Auto Summarizaion is an applied technology by using computer self-compile text summary.It is one of the major research directions on Natural Language Processing for Deep Learning and supports various application scenarios such as auto-generates reports,headline-generated,search preview for google,etc.At present,Auto Summarizaion doesn't work exactly like artificial text summarize which works in the way of understand the article before summarizing.Text dataset can be expressed by term vectors with Deep Learning thus the computer can learn to sumarize in neural networks.In the thesis abstract is a idea of finding out the key sentences and sentence keyword in the source document.This thesis combines some technologies such as sequence-to-sequence model,attention mechanism,builds an improving dual attention mechanism-Seq2 Seq model which based on both sentence and word level,to sumarize chinese text automatically.This thesis chooses LCSTS that comes from Sina weibo of HIT(Harbin Institute of Technology)and builds a short text dataset of a small Chinese report text.They are tested both on classical Seq2 Seq model and an improving dual attention mechanism-Seq2 Seq model by adopting intrinsic methods to estimate the result which mainly based on these models.The experimental results show that this improving dual attention mechanismSeq2 Seq model can improve the quality of text summary.