使用有序词语移动距离特征进行中文文本蕴含识别
发布时间：2018-10-05 07:10
【摘要】：提出了一种基于有序词语移动距离的中文文本蕴含识别方法,该方法基于word2vec词向量计算有序词语移动距离特征,进而利用有序词语移动距离特征和传统语言学特征通过支持向量机生成分类模型,然后使用分类模型进行蕴含识别,最终得到蕴含结果.该方法在RITE-VAL评测任务的CS数据上的Macro F1为0.629,超过RITEVAL的最优评测结果(BUPTTeam,0.615).实验结果表明,该方法可以提升中文文本蕴含识别系统的性能.
[Abstract]:A Chinese text implication recognition method based on the moving distance of ordered words is proposed. The method calculates the moving distance of ordered words based on word2vec word vector. Then, the classification model is generated by support vector machine (SVM) using the moving distance feature of ordered words and the traditional linguistic feature, and then the implication recognition is carried out by using the classification model, and the implication result is finally obtained. The Macro F1 on the CS data of RITE-VAL evaluation task is 0.629, which is higher than the optimal evaluation result (BUPTTeam,0.615) of RITEVAL. Experimental results show that this method can improve the performance of Chinese text implication recognition system.
【作者单位】： 北京邮电大学计算机学院;
【基金】：国家自然科学基金项目(U1536121,61370195)
【分类号】：TP391.1
，
本文编号：2252477
[Abstract]:A Chinese text implication recognition method based on the moving distance of ordered words is proposed. The method calculates the moving distance of ordered words based on word2vec word vector. Then, the classification model is generated by support vector machine (SVM) using the moving distance feature of ordered words and the traditional linguistic feature, and then the implication recognition is carried out by using the classification model, and the implication result is finally obtained. The Macro F1 on the CS data of RITE-VAL evaluation task is 0.629, which is higher than the optimal evaluation result (BUPTTeam,0.615) of RITEVAL. Experimental results show that this method can improve the performance of Chinese text implication recognition system.
【作者单位】： 北京邮电大学计算机学院;
【基金】：国家自然科学基金项目(U1536121,61370195)
【分类号】：TP391.1
，
本文编号：2252477
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2252477.html
