摘要:
提出了一种基于随机网络的在线评论情绪倾向性分类模型SCP-X (Shortest covering path-X). 首先引入了一种增量式创建词语顺序共现随机网络的方法, 并基于此随机网络以及情绪词表, 提出了一种基于评论序列最短覆盖路径(Shortest covering path, SCP)的情绪倾向性分类方法. 该方法具有以下两个优点: 1)能够对相对短小、随意性 较强、完整性较差的评论文本展开词语联想, 从而对完整性较差的评论数据进行属性值扩展; 2) 能够对评论文本的冗余属性进行约简, 约简后数据的属性规模为一般VSM模型 的10%左右. 本文最后设计了一组实验, 对以下算法进行了对比测试: TC, SVM, SCP-TC, SCP-SVM, SCP-HMM, SCP-Bayes. 结果表明本文提出的SCP-X方法对在线评论文本的倾向性分类效果更佳.
Abstract:
We propose a new method of sentiment classification named SCP-X (shortest covering path-X) for online comment based on the random network theory. A new approach which is proved to be effective by experiments is presented to create the word co-occurrenced network incrementally. With the network, the sequences of online comments, which are shorter, more optional and more fragmentary, are extended by shortest covering path (SCP) proposed in this paper. Using this algorithm, the amount of attributes is reduced to about 10% compared to VSM. Finally, experiments are designed to compare the results of the algorithms such as TC, SVM, SCP-TC, SCP-SVM, SCP-HMM, and SCP-Bayes. The results indicate that SCP-X is remarkably effective to classify online comments by sentiment orientation.
点击查看大图
下载:
