Text summarization resolves the issue of capturing essential information from a large volume of text data. Existing methods either depend on the end-to-end models or hand-crafted preprocessing steps. In this study, we propose an entity-centric summarization method which extracts named entities and produces a small graph with a dependency parser. To extract entities, we employ well-known pre-trained language models. After generating the graph, we perform the summarization by ranking entities using the harmonic centrality algorithm. Experiments illustrate that we outperform the state-of-the-art unsupervised learning baselines by improving the performance more than 10% for ROUGE-1 and more than 50% for ROUGE-2 scores. Moreover, we achieve comparable results to recent end-to-end models.

文本摘要解决了从大量文本数据中捕获基本信息的问题。现有方法取决于端到端模型或手工制作的预处理步骤。在这项研究中，我们提出了一种以实体为中心的汇总方法，该方法提取命名的实体并生成一个带有依赖解析器的小图。为了提取实体，我们采用了众所周知的预训练语言模型。生成图形后，我们通过使用谐波中心度算法对实体进行排名来进行汇总。实验表明，通过将ROUGE-1的性能提高超过10％，将ROUGE-2的得分提高50％以上，我们的性能优于最新的无监督学习基准。而且，我们获得了与最新端到端模型可比的结果。