Integrating Extractive and Abstractive Models for Long Text Summarization

With the explosive growth of information on the Internet, it becomes more and more important to improve the efficiency of information acquisition. Automatic text summarization provides a good means for quick acquisition of information through compression and refinement. While existing methods for automatic text summarization achieve elegant performance on short sequences, however, they are facing the challenges of low efficiency and accuracy when dealing with long text. In this paper, we present a twophase approach towards long text summarization, namely, EA-LTS. In the extraction phase, it conceives a hybrid sentence similarity measure by combining sentence vector and Levenshtein distance, and integrates it into graph model to extract key sentences. In the abstraction phase, it constructs a recurrent neural network based encoder-decoder, and devises pointer and attention mechanisms to generate summaries. We test our model on a real-life long text corpora, collected from sina.com, experimental results verify the accuracy and validity of the proposed method, which is demonstrated to be superior to state-of-the-art methods.