基于句子相似度的自动文摘评价方法

评价是自动文摘研究领域中非常重要的一个环节。针对目前自动文摘内部评价方法存在主观性强,无法体现深层语义的缺陷,提出了一种基于句子相似度的自动文摘评价方法。该方法是利用机器摘要和理想文摘中的句子相似度衡量文摘系统的性能,而不是单纯地利用词频或语句包含关系。实验表明,基于句子相似度的评价方法是一种更接近自然模型的评价方法。

The Evaluation Method for Automatic Summarization Based on Sentence Similarity

Evaluation is important for automatic summarization.The existing evaluation method has its disadvantages which include subjectivity and lack of deep semantic information.We proposed a method based on sentence similarity in this paper,which mainly took use of the difference between automatic summarization and standard summarization to evaluate the performance of the automatic summarization system.Study showed that it was closer to the natural evaluation method.