针对文本水印摘要攻击的语义损失量评估方法

摘要

为了控制文本水印自动摘要攻击造成的语义信息损失,在已有的自动摘要评估方法的基础上,针对自动摘要文本水印攻击算法提出一种评估文本语义损失度的算法.该方法通过量化句子语义,合理计算摘要攻击造成的语义损失；并分析了攻击造成语义损失的主要因素,以及这些因素和语义损失量之间的数学关系.该方法能够从语义信息损失的角度更真实地评测摘要攻击算法的失真度,实现了自动化评估.%Summary attack on natural language watermarking was likely to cause excessive loss of the texts' original meaning. In order to control the loss, this paper proposed a method based on the existing evaluation method of automatic summarization. It quantized the meaning of every sentences, then evaluated the meaning loss which caused by the summary attack. It analyzed the major factors of meaning loss, as well as the relationship between these factors and the meaning loss. This method gave the attack algorithm' s distortion a more realistic evaluation and was an automatic evaluation method.