基于语义规则增强的蒙古语情感分布学习  

Semantic rule enhancement based Mongolian emotion distribution learning

作　　者：杨蕾 苏依拉[1] 仁庆道尔吉[1] 吉亚图 乌尼尔[1] 路敏 YANG Lei;SU Yi-la;RENQING Dao-er-ji;JI Ya-tu;WU Ni-er;LU Min(College of Information Engineering,Inner Mongolia University of Technology,Hohhot 010080,China)

机构地区：[1]内蒙古工业大学信息工程学院,内蒙古呼和浩特010080

出　　处：《计算机工程与设计》2024年第7期2082-2089,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61966028、61966027);内蒙古自然科学基金项目(2021MS06028);内蒙古自治区攻关基金项目(2021GG0329)。

摘　　要：为完善基于先验知识的标记增强方法对于情绪信息的捕捉,提出一种基于语义规则增强的蒙古语情感分布学习方法(semantic rule enhancement based Mongolian emotion distribution learning, SRE-MEDL)。在情感轮和情感词典的基础上,引入程度词典和否定词典,得到各种情感词组合,以此制定相应的语义规则计算情感词权重,将其融入到标记增强中。在情感分布学习中融入从情感分布空间到实例特征空间的反向重构映射来弥补正向映射引起的原始信息丢失问题。对比实验结果显示,在蒙古语和中英文常用数据集上,SRE-MEDL方法在标记增强任务和情感分布学习中的表现均优于现有方法。To improve the emotional information capture of label enhancement method based on prior knowledge,a method of semantic rule enhancement based Mongolian emotion distribution learning was proposed.On the basis of emotion wheel and emotion dictionary,the degree dictionary and negative dictionary were introduced to obtain various emotion word combinations,so as to formulate corresponding semantic rules to calculate the weight of emotion words,which were incorporated into the label enhancement.The reverse reconstruction mapped from the emotion distribution space to the instance feature space was incorporated into the emotion distribution learning to compensate for the loss of original information caused by the positive mapping.Comparative experimental results show that the SRE-MEDL method outperforms existing methods in both label enhancement tasks and emotion distribution learning on commonly used datasets in Mongolian,Chinese,and English.

关 键 词：标记增强 语义规则 程度词 否定词 情感轮 蒙古语 情感分布学习 反向重构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]