基于句子平衡策略的动态遮盖文本摘要方法  

Dynamic Masked Text Summarization Method Based on Sentence Balancing Strategy

作　　者：雷罡 梁琨[1] 黄越洋 张贤坤[1] LEI Gang;LIANG Kun;HUANG Yueyang;ZHANG Xiankun(College of Artificial Intelligence,Tianjin University of Science and Technology,Tianjin 300457,China)

机构地区：[1]天津科技大学人工智能学院,天津300457

出　　处：《天津科技大学学报》2025年第1期49-56,共8页Journal of Tianjin University of Science & Technology

基　　金：国家自然科学基金项目(62377036)。

摘　　要：自动文本摘要技术是自然语言处理领域的焦点研究方向之一。针对现有特征提取方法捕捉语义能力较弱以及摘要结果中存在大量信息冗余的情况,本文通过调整句子嵌入阶段生成词向量的方式和遮盖策略,同时在句子选择阶段引入了冗余性、关键性联合评分机制,分别在中文短文本和长文本数据集上进行抽取式摘要的实验。实验结果表明,通过在句子嵌入和选择阶段进行一系列改进,本文模型的性能明显优于TextRank+BERT、BERTSUM等其他基准模型。Automatic text summarization technology is one of the focal research directions in the field of natural language processing.In response to the weak ability of existing feature extraction methods to capture semantics and the presence of a large amount of information redundancy in summary results,in our present study by adjusting the way of generating word vectors and the masking strategy during the sentence embedding phase,and introducing a joint evaluation mechanism of redundancy and criticality in the sentence selection stage,and experiments on extracted summaries were conducted on Chinese short text and long text datasets,respectively.The experimental results showed that by making a series of improvements in the sentence embedding and selection stages,the performance of our model was significantly better than other benchmark models such as TextRank+BERT,BERTSUM,etc.

关 键 词：自然语言处理 文本摘要 抽取式摘要 BERT 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]