微博客基本社会情绪的测量及效度检验

Weibo Social Moods Measurement and Validation

微博客积累的海量信息为直接快速地测量社会情绪提供了可能。本研究构建了微博客基本情绪词库,结合在线文本词汇匹配技术对数百万用户的情绪进行分析,得到了快乐、悲伤、愤怒、恐惧和厌恶五种基本社会情绪。发现快乐与其它四种情绪显著负相关,而四种情绪之间正相关,符合情绪维度理论;工作日的快乐情绪显著低于周末;重要节日和事件引起了社会情绪的相应波动。这些结果都表明基于微博客的社会情绪测量是有效的。

Weibo is an increasingly popular form of social media and accumulates vast amounts of information, making the measurement of social mood easily. The paper is about how to measure public mood using Weibo directly and efficiently. We proceed in three phases to measure and validate the social mood on Weibo. In the first phase, we create the Weibo Five Basic Mood Lexicon （Weibo-5BML）. In perspective of emotional categorical approach, there are five basic emotions including Happiness, Sadness, Fear, Anger and Disgust. We collect emotional words as many as possible and ask three psychological graduates to categorize every word. At last, we get the formal version of the Weibo-5BML. There are 818 emotional terms in the Weibo-5BML, in which Happiness has 306 terms, Sadness has 205 terms, Fear has 72 terms, Anger has 93 terms, and Disgust has 142 terms. In the second phase, we generate social mood time series. We analyze minute texts in Sina Weibo using a transparent approach named term-based matching technique, which matches the emotional terms used in each tweet against Weibo-5BML. The Weibo-5BML can capture a variety of naturally occurring emotional terms in Weibo tweets and map them to their respective social mood dimensions. The score of each basic mood dimension is thus determined as the sum of each tweet term that matches the Weibo-5BML each day. Then we obtain five basic social mood daily series from July 1, 2011 to November 30, 2012. In the third phase, we validate the Weibo social moods by different kinds of methods. First, we calculate the frequency of each social mood and find the frequency of happiness is higher than the other four social moods which is consistent to the relevant research of people expressing happiness more and the hyper-personal interaction model. Second, we calculate the correlation of five social moods and the result is consistent with the circumplex model of emotion. Happiness is negatively correlated with the other four kind of social moods, while the four kinds of social moods are positively correlated with each other. Third, we get the fluctuation of five social moods during a week and the result is similar to other relevant research. People are happier on weekends than workdays and the unhappiest day is Wednesday. At last, we match the five basic Weibo social moods against the fluctuations recorded by major events of social and popular culture and find these events cause corresponding fluctuation in Weibo social mood. For example, people are happy on both Chinese and Western holiday and the public are angry because of the conflict of Diaoyu Island. People are sad about the fragile life and the dead or injured passengers at the beginning of ＂7.23 Wenzhou Train Collision＂, and are angry at the fourth and fifth day because of the cause tracing. All of these results indicate that the social mood on Weibo is effective on capturing the public＇s mood. It is useful for combining the psychological theory and techniques of computing science and these text and image information on Internet provide the valuable resources and opportunities for researchers to study the individual or collective characteristics.