Research And Realization Of Graph-based Subjective Multi-document Automatic Summarization

With the rapid development of Internet technique, It has entered the Big Data Age. There are many same-topic documents in the Internet. These documents contain the same content and the different content. People hope that they spend a minimum of time gaining more information from the same-topic documents. Automatic summarization has become a hotspot for natural language processing.This thesis introduces a graph-based subjective multi-document automatic summarization's method. The aim is to extract some sentences from the same-topic subjective multi-document and generate a text summarization. The general summarization is aimed at the objective documents, which aren't related to the tendency of semantics. The author hoped to find an effective and accurate method to select some sentences from the subjective texts. Thus, the author considered that the multi-document summarization's content is both concise and comprehensive. In this thesis, the author used the graph-based unsupervised learning and the semi-supervised learning which judged sentence semantic tendency. Unsupervised learning generates summarization without training on a labeled set of documents. There is no bias problem in the summarization because it doesn't need human annotation. Semi-supervised learning judges the sentence semantic tendency with a small amount of labeled data.We improve the method of Lex Rank and GRASSHOPPER. Testing the improved methods and compared with testing the original methods, the improved methods' testing results have been significantly better. This indicates the improved methods have been effective.