信息安全研究 ›› 2019, Vol. 5 ›› Issue (11): 981-987.

• 内容安全与人工智能专题 • 上一篇    下一篇

智能人机对话中的负面情感检测

罗观柱,赵妍妍,秦兵,刘挺   

  1. 哈尔滨工业大学计算机科学与技术学院
  • 收稿日期:2019-11-08 出版日期:2019-11-15 发布日期:2019-11-20
  • 通讯作者: 罗观柱
  • 作者简介:罗观柱, 1991年5月出生,硕士研究生,主要研究领域为情感分析等,E-mail: gzluo@ir.hit.edu.cn. 赵妍妍,1983年生,博士,副教授,博士生导师,主要研究领域为社会媒体计算等,E-mail: yyzhao1983@126.com. 秦兵,1968年生,博士,教授,博士生导师,主要研究领域为自然语言处理、文本挖掘等,E-mail: qinb@ir.hit.edu.cn. 刘挺,1972年生,博士,教授,博士生导师,主要研究领域为自然语言处理、社会媒体计算等,E-mail: tliu72@foxmail.com.

Negative Sentiment Detection in Intelligent HumanMachine Dialogue

  • Received:2019-11-08 Online:2019-11-15 Published:2019-11-20

摘要: 人工智能使得智能人机对话系统的应用日趋广泛,但对话系统中常常含有低俗、暴力、谣言等负面情感信息,给对话系统造成了不良的影响,因此检测对话系统中的负面情感显得至关重要.针对智能人机对话系统中的负面情感内容,提出了一种基于深度学习的负面情感检测模型,该模型利用预训练词向量和BiLSTM可以有效地捕捉文本语义与上下文信息.相对于传统的词典匹配算法,大大减少了对词典的依赖程度,能够智能地识别相似的负面情感表达形式,可以更加有效地检测对话系统中的负面情感,在净化对话系统中起到了重要作用.

关键词: 人机对话, 负面情感, 情感分析, 低俗检测, 深度学习

Abstract: Artificial intelligence makes the application of intelligent humanmachine dialogue system more and more widely, but the dialogue system often contains negative sentiment information such as vulgarity, violence and rumors, which has a bad influence on the dialogue system. Therefore, it is very important to detect negative sentiment in the dialogue system. Aiming at the negative emotion content in the intelligent humanmachine dialogue system, a negative sentiment detection model based on deep learning is proposed. This model can effectively capture text semantics and context information by using pretraining word vector and BiLSTM. Compared with the traditional dictionary matching algorithm, it greatly reduces the dependence on the dictionary, and also can intelligently identify similar negative sentiment expression forms, which makes detecting negative sentiment in the dialogue system more effectively and plays an important role in purifying the dialogue system.

Key words: human-machine dialogue, negative sentiment, sentiment analysis, vulgarity detection, deep learning