Multi-Document Extractive Text Summarization via Deep Learning Approach

Today, given the huge amount of information, summarization has become one of the most applicable topics in data mining that can help users gain access to useful data over a short period of time. In this study, two multi-document extractive text Summarization systems are introduced. The major objective of this research is to use autoencoder neural network and deep belief network separately for scoring sentences in a document to compare their performances. Deep neural networks can improve the results by generating new features. The abovementioned systems were tested on DUC 2007 dataset and evaluated using ROUGE-1 and ROUGE-2 criteria. The results show a better performance of autoencoder network versus deep belief network. It is also possible to compare these values with results of other systems to realize the effectiveness of the proposed methods.