多知识源融合的自动摘要系统研究与实现  

Research and design of summarization system based on multi-knowledge sources fusion

作　　者：陈燕敏[1] 王晓龙[1] 刘秉权[1] 楼喜中[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《高技术通讯》2006年第4期337-341,共5页Chinese High Technology Letters

基　　金：国家自然科学基金（60435020）和国家863计划（2002AA117010-09）资助项目.

摘　　要：提出并实现了一种能有效地融合改进的浅层主题特征分析方法、词汇链方法、话语结构方法的分析结果来生成文摘的多知识源融合的自动摘要系统，并对其进行了评测实验。评测结果表明，该系统在捕获文章特征的同时较好地保持了原文的内容及内在逻辑结构，生成的文摘具有良好的连贯性和流畅性；与采用单一方法的自动摘要系统相比较，生成的文摘质量有明显提高。An automatic text summarization system based on fusing multiple knowledge sources is proposed. It generates summaries by fusing the results produced by the improved method based on add terms, the improved method based on lexical chains and the improved method based on discourse structure. Evaluation results show that the presented system not only captures the characteristics of texts but also retains the content and inherent logic of sources, and high-quality summaries are produced from arbitrary Chinese texts. The proposed system is compared to systems based on single method and it is shown that it produces either comparable or better summaries overall.

关 键 词：自动摘要 特征词 词汇链 话语结构 融合 

分 类 号：N94[自然科学总论—系统科学] G20[文化科学—传播学]