基于主题模型的半监督网络文本情感分类研究

针对网络评论文本的情感分类问题中存在的数据的不平衡性、无标记性和不规范性问题,提出一种基于主题的闽值调整的半监督学习模型,通过从非结构化文本中提取主题特征,对少量标注情感的文本训练分类器并优化指标调整闽值,达到识别用户评论的情感倾向的目的。仿真研究证明阈值调整的半监督模型对数据非平衡性和无标记性具有较强的适应能力。在实证研究中,对酒店评论文本数据构建的文本情感分类器显示该模型可以有效预测少数类评论样本的情感极性,证实了基于主题模型的闽值调整半监督网络评论文本情感分类模型在实际问题中的适用性与可行性。

Study on Semi-Supervised Sentiment Classification of Web Context Based on Topic Model

The study on the sentiment classification is challenged by the imbalanced,unmarked and nonstandard web context data.In this paper,we proposes an adaptive semi-supervised topic-based classifier to figure the above issues.Numerical study shows that the proposed method has strong adaptability to the imbalanced,unmarked datasets.A sentiment classification of hotel comment context gains effectiveness in predicting sentimental polarity of minority group in real study,which has confirmed the applicability and feasibility of this adaptive semi-supervised topic-based classifier in practical problems.