提出了基于知网概念特征的文本综述方法,探讨了语句相似度计算、主题区域发现、新颖度获取和综述生成等关键技术.通过对知网的改造,获取了关键词的概念特征,实现了同义词概念扩充,在文档语义相关性基础上,实现了多文档的自动综述.采用一种基于综合评价理论的文本综述评价方法,从综述的表达质量、表述内容和基于Q8LA的信息性评价三个方面实现了对综述的评价.实验结果表明该方法有效可行.
An approach of multi-document summarization based on How-net is presented. Several key techniques are deeply discussed including sentence similarity computation, topic information identification, topic novelty and summarization generation. On the basis of concept feature derived from improved How-net, the multi-document automatic summarization has been generated. The summarization evaluation method based on integration evaluation is also developed, and it gives a systematic evaluation about quality, content and information evaluation based on Q&A. The experiment results show that the method is effective.