TP-AS :A Two-phase Approach to Long Text Automatic Summarization

Shuai WANG,Xiang ZHAO,Bo LI,Bin GE,Daquan TANG
DOI: https://doi.org/10.3969/j.issn.1003-0077.2018.06.009
2018-01-01
Abstract:With the explosive growth of information on the Internet ,it becomes more important to improve the effi-ciency of knowledge acquisition.Automatic text summarization techniques provide a good means for fast knowledge acquisition by compressing and refining information.Existing automatic text summarization methods ,when dealing with long text ,exhibit poor accuracy ,and fail to meet users' need for performance.In this paper ,we propose a two-phase automatic summarization method for long text ,namely ,TP-AS .Firstly ,it employs a hybrid semantic similarity computation method based on a graph model to extract key sentences .Then ,it constructs a recurrent neural network encoder-decoder model with attention and pointer mechanisms to generate summaries .Through experiments on real large-scale long-text corpora in financial area ,the effectiveness of TP-AS is verified ,and its accuracy for automatic summarization notably outperforms other existing methods .
What problem does this paper attempt to address?