基于微博的细粒度情感分析

发布时间：2018-12-12 08:06

【摘要】：【目的】对微博进行细粒度情感分析,将情感分为8类,并计算其情感强度值,从而尽可能还原微博用户情感。【方法】通过微博语料分析构建疑问词词表,在大连理工大学情感词汇本体DUTIR的7类情感基础上,丰富一类情感"疑",并利用点互信息法构建表情符号词典,还综合考虑否定词和程度副词对情感表达的影响,利用Python从新浪微博上获取数据,并用R语言的jieba R包进行分词,对情感进行分类并计算其强度。【结果】得到微博用户对于糖尿病7类常用药物的8类情感占比及情感强度,并通过正确率、召回率、F值对结果进行验证,其中"怒"和"哀"的正确率最高,分别为85.73%和83.05%,而"乐"和"好"的召回率与F值均最高,为81%以上。本文新增情感"疑"的正确率、召回率、F值分别为77.33%、78.58%、77.95%,均值在8类情感中排名前列,说明其情感识别较好。【局限】由于本文依赖于情感词典进行情感分析,因此为了更好的分析结果,情感词典仍需进一步完善。【结论】本方法具有较高的识别率和可靠性,能够更好地对微博上的情感分类进行细粒度分析。
[Abstract]:[objective] to analyze the fine granularity emotion of Weibo, divide the emotion into 8 categories, and calculate the emotion intensity value, so as to restore the user emotion as much as possible. [methods] the lexicon of interrogative words was constructed by analyzing the data of Weibo. On the basis of seven kinds of emotion of DUTIR, Dalian University of Technology, this paper enriches a kind of emotion "doubt", constructs an emoji dictionary with the method of point mutual information, and synthetically considers the influence of negative words and degree adverbs on emotion expression. Python was used to obtain data from Sina Weibo, and jieba R package in R language was used to classify emotion and calculate its intensity. [results] eight kinds of emotion and emotional intensity of Weibo users for 7 kinds of commonly used drugs for diabetes mellitus were obtained. The correct rate of "anger" and "sorrow" were 85.73% and 83.05, respectively, while the recall rate of "Yue" and "good" and F value were the highest (above 81%). In this paper, the correct rate, recall rate and F value of the new affective "suspicion" are 77.33 and 78.58, respectively. It shows that the emotion recognition is better. [limitation] because this paper relies on emotion dictionary for emotion analysis, in order to better analyze the result, the emotion dictionary still needs to be further improved. [conclusion] this method has high recognition rate and reliability. Can better carry on fine-grained analysis to the emotion classification on Weibo.
【作者单位】：吉林大学公共卫生学院;大连理工大学中日国际信息与软件学院;
【分类号】：TP391.1

【相似文献】