基于复杂网络的情感分类特征选择

发布时间：2018-03-01 23:05

本文关键词： 复杂网络特征选择情感分类情感词典　出处：《计算机应用研究》2017年04期 　论文类型：期刊论文

【摘要】：首先针对公共情感词典对专业领域适用性较低的问题,以公共情感词典作为种子情感词典,以评论语料库中未出现在公共情感词典中的形容词作为候选情感词,在此基础之上利用点互信息理论构建专业领域的情感词典;其次针对在线评论情感分类问题,利用复杂网络理论提出了一种新的情感分类特征选择算法,改进了传统特征选择算法忽略特征语义相关信息、遗漏评论情感资源的问题。通过构建候选特征词关系网络,利用复杂网络节点重要性理论,考虑节点的局部和全局重要性,提出了利用网络节点的度中心性、介数中心性和接近中心性综合衡量节点重要性来选择情感分类特征的算法NTFS(complex network feature selection)。最后以i Phone手机的在线评论为实验数据,利用SVM、NNET、NB分类器对比了NTFS、GI、CHI传统特征选择方法,实验证明NTFS在分类性能上优于GI、CHI算法。
[Abstract]:Firstly, aiming at the low applicability of the public emotion dictionary to the professional field, the public emotion dictionary is used as the seed emotion dictionary, and the adjectives which are not found in the public emotion dictionary in the comment corpus are used as the candidate affective words. On the basis of this, using the point mutual information theory to construct the emotion dictionary of professional field. Secondly, aiming at the online comment emotion classification problem, a new emotion classification feature selection algorithm based on complex network theory is proposed. This paper improves the traditional feature selection algorithm which neglects the feature semantic information and omits the comment on emotional resources. By constructing the candidate feature relational network and using the theory of node importance of complex network, the local and global importance of nodes is considered. In this paper, an algorithm, NTFS(complex network feature selection, is proposed to measure the importance of nodes by using the degree centrality, medium centrality and proximity centrality of network nodes. Finally, the online comment of I Phone mobile phone is used as the experimental data. The traditional feature selection method of NTFS is compared with the traditional feature selection method by using SVMNNETN NB classifier. The experimental results show that NTFS is superior to the GIZCHI algorithm in classification performance.
【作者单位】：大连理工大学系统工程研究所;
【基金】：国家自然科学基金面上项目(61471083) 国家教育部人文社科研究规划基金资助项目(14YJA630044)
【分类号】：TP391.1;O157.5

【相似文献】