基于改进TextRank的抽取式自动文本摘要生成方法  

Extraction Automatic Text Summarization Generation Method Based on Improved TextRank

作　　者：梁高鹏 徐鲁强[1] LIANG Gaopeng;XU Luqiang(School of Computer Science and Technology,Southwest University of Science and Technology,Mianyang 621010)

机构地区：[1]西南科技大学计算机科学与技术学院,绵阳621010

出　　处：《计算机与数字工程》2024年第12期3643-3648,共6页Computer & Digital Engineering

摘　　要：模型TextRank在抽取式自动文摘方法中的表现相对较好,但其在初始文本质量和节点权重得分计算等环节,仍有较大的调整提升空间。针对此情况,提出了一种新的调整方法。结合自动文摘的实际应用环境与文本在文学方面的表达特点,通过在文本预处理阶段增加预排序流程来突出表现文本的主旨观点,减少语义重复内容,提升输入文本的质量。通过对相似度计算公式的调整,在最后的节点得分公式中将词频、与标题相似度、段间位置等因素按照特定的比例加入到权重系数中参与得分计算来优化整个计算流程。最终的实验结果表明,调整后的模型在各方面的得分情况要优于原模型,生成的摘要质量更高,更接近于人工生成的摘要。TextRank,an automatic text summarization model,performs relatively well in the extraction automatic summariza-tion method,but there is still much room for adjustment and improvement in the two links of initial text quality and node weight score calculation.In view of this situation,a new adjustment method is proposed.Combined with the practical application environ-ment of automatic summarization and the literary expression characteristics of the text,the main idea of the text is highlighted and the quality of the input text is improved by adding a pre-sorting process in the text preprocessing stage.By adjusting the similarity calculation formula,factors such as word frequency,similarity with title,and position between segments are added to the weight co-efficient according to a specific proportion in the final node score formula to participate in the score calculation,so as to optimize the whole calculation process.The final experimental results show that the adjusted model is better than the original model in all as-pects.

关 键 词：抽取式 自动文本摘要 TextRank 预排序 权重得分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]