摘要
提出了一种基于LDA-Col模型的无监督主题情感混合(UTSU)模型。采用词序流对文本进行表示,对每个句子采样情感标签,对每个词采样主题标签,得到文本的主题情感分布。这种采样方式既符合语言的情感表达,又不会缩小词之间的主题联系,克服了ASUM模型和JST模型在同一层盘子中采样主题标签和情感标签的缺陷。实验表明,UTSU模型的情感分类性能比有监督的情感分类方法稍差,但在无监督的情感分类方法中效果最好,情感分类综合指标比ASUM模型提高了3%,比JST模型提高了17%。
An unsupervised topic and sentiment unification model,UTSU model for short,is proposed based on the LDA-Col model.Unlike the ASUM model and the JST model that sample sentiments and topics from the same plate,the UTSU model imposes the constraint that all words in a sentence are generated from one sentiment and each word in the sentence is generated from one topic.The constraint accords with the sentiment expression of language and will not limit the topic relation of words.The experiments of sentiment classification show that the result of the UTSU model is close to the results of supervised classification methods and outperforms other topic and sentiment unification models.Comparisons with the ASUM and JST models show that the UTSU model improves the F1 value of sentiment classification by about 3% and 17%,respectively.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2013年第1期120-125,共6页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(611100042)
关键词
文本情感分类
无监督学习
混合模型
text sentiment classification
unsupervised learning
unification model