摘要
通过分析兵棋演习过程中的常见问题,设计了一个针对兵棋演习特殊情景的问句分类模型。问句分类模型基于统计方法,利用Word2vec工具生成词向量,利用TextRank算法结合IDF值来生成词权重,共同完成问句表征。并综合考虑算法复杂度以及问句相似度计算的精确度,通过两个不同的问句相似度模型,以及改进的KNN算法来实现最终的问句分类。WMD(Word Mover's Distance)算法是在词向量基础上计算问句相似度较为精确的算法,但同时存在算法复杂度过高的缺点,论文通过改进的KNN算法将其与传统算法结合,来更好地完成需要的问句分类任务。
By analyzing the common problems in the process of war gaming,a question classification model is designed for aQA(question answering)system oriented a specific situation. The question classification model generates word vectors by word2 vecbased on statistical methods,and generates word weight by TextRank algorithm so as to complete the question representation. Theclassification combines two different models of question similarity calculation through the improved KNN(K Nearest Neighbor)algo-rithm,balancing the computation complexity and accuracy. The WMD(Word Move Distance)algorithm is based on the word vectorto calculate the similarity of questions more accurate algorithm,which also has the disadvantage of high algorithm complexity howev-er. In this paper,the improved KNN algorithm is combined with the traditional algorithm,in order to complete the required questionclassification task better.
作者
孙泽健
司光亚
刘洋
SUN Zejian;SI Guangya;LIU Yang(Department of Information Operation & Command Training,NDU,Beijing 100091)
出处
《计算机与数字工程》
2019年第2期308-313,319,共7页
Computer & Digital Engineering
基金
国家自然科学基金军民共用重大研究计划联合基金"基于仿真大数据的武器装备体系复杂性机理及效能评估方法研究"(编号:U1435218)
国家自然科学基金青年基金"基于仿真大数据的信息化作战体系分析方法研究"(编号:61403401)资助
