节点文献
中文自动文摘系统的综合评价模式
Evaluation of Chinese Text Automatic Abstracting System Based on Integration Evaluating Method
【Author】 Lu Ye Lin Hongfei Zhao Jing Dept. of Computer Science and Engineering, Dalian University of Technology, Dalian 116024
【机构】 大连理工大学计算机科学技术系;
【摘要】 采用五种评测方法进行了文摘评价实验:文摘的可接受性评价、与“理想”文摘的比较、基于Q&A的文摘信息性评价、文摘的关联性评价和基于分类的评价。不仅验证了文摘系统所采用的技术的性能,更重要的是分析了影响文摘评价质量的因素。试验结果表明:不同参数对每一项实验的评价结果都有很大的影响。根据各项评价方法的特点,旨在发挥各项评价方法的优势,使用了一种基于综合评价理论的综合性评价方法,采用了线性加权模型,实现了对文摘系统的公正、全面、系统的评价。
【Abstract】 Five methods were used for evaluation of automatic abstracting systems: an evaluation of the readability and acceptability, an evaluation of generated abstracts against an "ideal" abstract, an evaluation of coverage of key concepts based on Q&A, an evaluation of relevance and an evaluation based on categorization. The performance of automatic abstracting system we had used was not only testified by all of these methods, but the factors affecting the quality of abstracting evaluation were also analyzed. The results show that different parameters of an experiment can dramatically affect how well a system scores. According to the characteristic and virtues of all methods, a synthetic evaluating method based on integration evaluating method was developed, linear weighted model was used, and a just, overall and systemic evaluation was consequently implemented.
【Key words】 readability; Q&A; relevance assessment; integration evaluating method; linear weighted model;
- 【会议录名称】 全国第八届计算语言学联合学术会议(JSCL-2005)论文集
- 【会议名称】全国第八届计算语言学联合学术会议(JSCL-2005)
- 【会议时间】2005-08
- 【会议地点】中国南京
- 【分类号】TP391.12
- 【主办单位】南京师范大学、清华大学智能技术与系统国家重点实验室