摘要
提出了一种基于题聚类的自动摘要算法.该算法在采用统计方法的同时.又适当结合知识理解,既摆脱了领域限制,也使摘要的结果更为准确.此外,为了能够全面反映信息样本的主要内容,而又不产生信息(?)余,本文提出的摘要算法还力图适应于不同的样本、动态确定摘要长度.为此.本文首先构造出新的互依赖模型,为摘要算法选择较为准确的属性.接着,挖掘出评估语句重要性的新规则.为摘要算法提供选择为重要语句的尺度.最后,提出了一种较为客观的、基于任务的摘要性能评估算法.
In this paper, an algorithm which automatically summarizes a document by extracting subtopics from the sentences is based on statistics and partially understanding knowledge, in order to get better summarization and get rid of the restriction of information domain. Resides, since it is diffcult to determine the length of summaries manually, the algorithm also strives to obtain a better summary with proper length. To this end, a new module of mutual dependence is put forward too and used to select features, which can selects accuracy features for the summarizing algorithm. And then new rules to evaluate sentences are brought forward. Furthermore, a new task-based algorithm to evaluate summarization impersonally is offered.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2004年第3期291-298,共8页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金(No.60173027)
关键词
自动摘要
互依赖
聚类
信息检索
Automatic Summarization
Mutual Dependence
Clustering
Information Retrieval