面向短文本的情感折射模型

发布时间：2018-01-28 05:01

本文关键词： 情感分析折射情感折射率情感词介质情感极性　出处：《情报学报》2017年02期 　论文类型：期刊论文

【摘要】：移动互联网时代产生海量的简短网络信息,快速、准确地获取这些数据中用户表达的情感信息具有广泛的应用场景。本文考虑人类按顺序阅读文本以获取情感信息的习惯,利用光的折射对此进行模拟,提出一种面向短文本分类的情感折射模型(Sentiment Refraction Model,SRM)。首先,从若干种子情感词出发,利用word2vector及k最近邻分类算法启发式地构建包含喜、怒、哀、乐、惧、恶六类情感的情感词典,认定每类情感词具有同等强度的情感,且这些情感词在不同的上下文中具有一定的情感折射率。其次,针对一条短文本,情感光线以给定的初始入射角向包含若干情感词的文本中传播,经过不同情感词介质的连续折射,情感光线传播方向产生相应的变化,通过情感光线出射角与初始入射角的差值即可判定文本的情感极性。最后,用NLPCC,COAE等公布的标准数据集对本文方法进行评测,分别与基于情感极性加权求和、朴素贝叶斯以及支持向量机分类方法进行对比。实验结果表明,情感折射模型在不同类型的短文本数据集上均有较好的表现,此外,针对简单词典与扩展词典的情感分类结果对比也证实了情感词典扩展方法的有效性。
[Abstract]:The era of mobile Internet produces a mass of short network information, fast. Accurate acquisition of the emotional information expressed by users in these data has a wide range of applications. This paper considers the habit of reading text in sequence to obtain emotional information and simulates it by the refraction of light. An emotion refraction model for short text classification is proposed. Firstly, some seed affective words are introduced. Using word2vector and k-nearest neighbor classification algorithm to construct an emotional dictionary which includes six kinds of emotions: joy, anger, sadness, joy, fear and evil, and finds that each affective word has the same intensity of emotion. And these words have certain emotional refractive index in different contexts. Secondly, for a short text, the emotional light propagates to the text containing a number of emotional words with a given initial incidence angle. Through the continuous refraction of different emotional words, the direction of emotional light propagation changes accordingly. The emotional polarity of the text can be determined by the difference between the initial incidence angle and the incident angle of emotional light. The method is evaluated with the standard data set published by NLPCC Con COAE et al., and the sum based on affective polarity is calculated respectively. Naive Bayes and SVM classification methods are compared. The experimental results show that the emotional refraction model has a good performance on different types of short text dataset, in addition. The comparison of affective classification results between simple dictionaries and extended dictionaries also proves the validity of affective dictionary expansion methods.
【作者单位】：北京邮电大学经济管理学院;北京信息科技大学网络文化与数字传播北京市重点实验室;
【基金】：国家自然科学基金项目(61271304) 教育部人文社会科学研究规划基金(16YJA630063)
【分类号】：TP391.1
【正文快照】： 移动互联网的快速发展进一步加剧了信息时代的数据膨胀。网民可以通过移动终端快速、便捷地参与信息发布,表达自己对人物、事件、产品等的观点和态度。这些评论大多产生于以微博、电子商务网站等为代表的在线社会网络中,趋向于简短化,口语化以及多样化。针对这些海量的用户生

【相似文献】