摘要
通用情感词典(GPEL)对情感词语所在上下文的背景建模方面表现较差。针对此问题,提出一种领域特定情感词典(Domain Specific Emotion Dictionary,DSED)生成方法的扩展方法。所提方法在特征提取中使用了DSED提供的知识,而非简单的词语计数。利用在训练文档上学习到的DSED知识,提取出基于词典的特征。词性标注、情绪词典和GPEL作为提取情感分类相关特征的外部资源。实验在SemEval-2007、微博数据集和博客数据集三个公开数据集上进行,实验结果表明,所提方法提取出的特征显著优于从GPEL提取出的特征,与逐点互信息(PMI)、n元语法等方法相比,所提方法的性能更优。
General purpose emotion lexicons(GPEL)performs poorly in the context modeling of emotion words.Aiming at this problem,we propose an extension method of domain specific emotion dictionary(DSED)generation method.The proposed method used the knowledge provided by DSED in feature extraction instead of simple word counting.The features based on dictionary were extracted by using the knowledge of DSED learned from training documents.Part of speech tagging,emotion dictionary and GPEL were taken as external resources to extract relevant features of emotion classification.The experiment was carried out on SemEval-2007,Weibo data set and blog data set three public data sets.The experimental results show that features extracted from the proposed method are significantly better than those extracted from GPEL.Compared with point-by-point mutual information(PMI),n-ary syntax and other methods,the proposed method has better performance.
作者
颜明阳
闫国梁
李明兰
Yan Mingyang;Yan Guoliang;Li Minglan(Primary Education College,Jining University,Qufu 273100,Shandong,China;China Telecom Post and Telecommunications Consulting Design Institute Co.,Ltd.,Beijing 100000,China;College of Mathematics and Statistics,Qingdao University,Qingdao 266071,Shandong,China)
出处
《计算机应用与软件》
北大核心
2022年第6期176-182,共7页
Computer Applications and Software
基金
山东省自然科学英才基金项目(ZR2015FM023)
山东省杰出青年科学基金项目(JQ201419)。
关键词
通用情感词典
领域特定情感词典
上下文
情感分类
相关特征
General purpose emotion lexicons
Domain specific emotion dictionary
Context
Sentiment classification
Relevant features
