电子学报 ›› 2015, Vol. 43 ›› Issue (9): 1875-1880.DOI: 10.3969/j.issn.0372-2112.2015.09.029

• 科研通信 • 上一篇    

基于LDA的多粒度主题情感混合模型

欧阳继红1,2, 刘燕辉1,2, 李熙铭1,2, 周晓堂1,2   

  1. 1. 吉林大学计算机科学与技术学院, 吉林长春 130012;
    2. 符号计算与知识工程教育部重点实验室, 吉林长春 130012
  • 收稿日期:2014-04-08 修回日期:2014-08-11 出版日期:2015-09-25
    • 作者简介:
    • 欧阳继红 女,1964年生于吉林长春,教授,博导,研究方向:知识工程与专家系统、空间推理和数据挖掘.E-mail:ouyj@jlu.edu.cn;刘燕辉 男,1989年生于山东德州,硕士,研究方向:文本分类与情感分析.E-mail:sdpy_lyh@163.com
    • 基金资助:
    • 国家自然科学基金 (No.61170092,No.61133011,No.61103091)

Multi-Grain Sentiment/Topic Model Based on LDA

OUYANG Ji-hong1,2, LIU Yan-hui1,2, LI Xi-ming1,2, ZHOU Xiao-tang1,2   

  1. 1. College of Computer Science and Technology, Jilin University, Changchun, Jilin 130012, China;
    2. Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, Changchun, Jilin 130012, China
  • Received:2014-04-08 Revised:2014-08-11 Online:2015-09-25 Published:2015-09-25
    • Supported by:
    • National Natural Science Foundation of China (No.61170092, No.61133011, No.61103091)

摘要:

主题情感混合模型(Reverse-Joint Sentiment/Topic Model;Joint Sentiment/Topic Model)能够有效地同时抽取文档的主题和情感信息,在情感分析领域受到广泛的关注,因为没有考虑整体分布与局部分布的关系,导致分类效果不佳且不稳定.本文同时考虑两个粒度上的情感/主题分布——文档级和局部,提出多粒度的主题情感混合模型(MG-R-JST;MG-JST).MG-R-JST/MG-JST在文档级分布和局部分布的共同作用下生成单词的情感/主题;使用吉布斯采样进行模型推理,并给出了推理过程;在MR与MDS数据集上进行实验,实验结果表明本文算法分类效果优于主题情感混合模型,且稳定性更好.

关键词: LDA, 主题情感混合模型, 情感分析, 多粒度

Abstract:

The topic and sentiment unification model (Reverse-Joint Sentiment/Topic Model;Joint Sentiment/Topic Model) can effectively extract information of topic and sentiment simultaneously and receives wide attention in the field of sentiment analysis,because it does not consider the relationship between the overall distribution and local distribution,so the classification performance is not good and stable.This paper proposed the multi-grain topic and sentiment unification model (MG-R-JST;MG-JST) by taking into account both grains on sentiment/topic distribution—document-level and local-level.MG-R-JST/MG-JST generated the sentiment/topic of words on the effect of the document-level and local-level distribution.we used Gibbs sampling for model inference and showed the process.Experiments on the dataset of MR and MDS demonstrate the effectiveness of the proposed method,and the classification performance is better and more stable than the topic and sentiment unification model.

Key words: LDA, topic and sentiment unification model, sentiment analysis, multi-grain

中图分类号: