Automatic Summarization Research for Long Texts Targeting Think Tanks

Think tanks are committed to researching social, political, economic, and other issues, providing relevant analyses and recommendations to help decision-makers make informed decisions. As the number of think tank reports continues to increase, how to quickly identify valuable information has become key to improving efficiency. Although automatic text summarization technology provides an effective way to achieve this goal, there is a common issue in think tank texts that the original text sequence is too long and leads to truncation when inputting into models. In response to this problem, this paper proposes a hybrid long text summarization model. Firstly, a BiLSTM-based extractive method is used to obtain summary candidate sentences. Then, a pre-training method based on T5-Pegasus is used to generate the final summary result. The results show that the long text summarization model outperforms other text summarization models in think tank text summarization tasks, with Rouge-1, Rouge-2, and Rouge-l values of 27.08, 15.63, and 25.43, respectively, demonstrating superior summarization performance on think tank texts.