基于主题加权LDA模型的情感分类方法

针对LDA(Latent Dirichlet Allocation)主题模型生成的大量topic,很大部分topic内部词语相关度很低,可解释性差,对语言模型后的应用效果带来一定的影响.针对这一问题,该文提出了一种基于主题加权LDA模型的情感分类方法,该模型实现不同主题中内部相关的词语特征加权计算,能够消除不同主题内具有相关度词语的相互影响.实验结果表明,与传统LDA模型分类方法对比,该文提出的基于主题加权LDA模型的情感分类方法平均F1值提高了6.7%～8.1%,验证了该文提出的方法是有效的,提高了分类效果.

Emotion Classification Method Based on Topic Weighted LDA Model

For the large number of topics generated by the LDA (Latent Dirichlet Allocation) theme model, the relevance of the internal words is very low, poor interpretation, and the effect of the language model is affected. In order to solve this problem, an emotion classification method based on topic weighted LDA model has been proposed in this paper, which can realize the weighting calculation of words in different themes, and can eliminate the influence of words with relevance in different themes. The experimental results show that compared with the traditional LDA model classification method, the average F1 value of the emotion classification method based on the topic weighted LDA model is improved by 6.7%-8.1%, which proves that our proposed method is effective and improved classification effect.