基于遗传算法的综合答案抽取及评价方法  

Comprehensive answer extraction and evaluation method based on genetic algorithm

作　　者：陈琛[1] 曹苏燕[2] 应捷[2] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210003 [2]南京交通职业技术学院电子信息工程系,江苏南京211188

出　　处：《现代电子技术》2013年第4期69-72,共4页Modern Electronics Technique

基　　金：国家重点基础研究发展规划(973)课题(2006AA01Z201)

摘　　要：在答案抽取过程中,需要对答案候选集按句子的相似度进行排序,抽取出相似度大于阈值的答案,为了提高答案抽取的各项指标并使之更均衡,提出了一种综合答案抽取和评价的方法。通过对句子的多个特征进行分析,在综合这些答案抽取算法的基础上,对答案的准确率和召回率的评价指标综合考量。实践证明遗传算法是一种简单有效的优化算法,应用遗传算法对句子相似度特征权重做优化,使权重的分配更加合理,从而计算结果达到最优。During the answer extraction, it is necessary to sort the answer candidate set by the similarity of sentences and extract the answers whose similarity is greater than the threshold value. In order to improve the each index of answer extraction and make it more balanced, a comprehensive answer extraction and evaluation method is put forward in this paper. A compre- hensive consideration to the evaluation index of accuracy rate and recall rate was given on the basis of multi-feature analysis of the sentences and synthesization of answer extraction algorithms. It has proved in practice that the genetic algorithm is a simple and effective optimization algorithm. The application of genetic algorithm in the optimization for feature weight of sentence simi- larity makes the distribution of weight more reasonable. Therefore, the optimal calculation results were obtained.

关 键 词：答案抽取 问答系统 句子相似度 遗传算法 

分 类 号：TN911.34[电子电信—通信与信息系统]