Due to the rapid development of the Internet, more and more social events can rapidly ferment and widely spread on the Internet. The social media has become an important way to dominate the trends of social events. It may even form a major social event that requires the government to respond promptly. The negative social events may lead to social instability once they are not properly dealt with. The pre-alarm system for the trends of social events can help the government know better about social conditions and public opinions, and can provide references on how to deal with it in the first place; thus, the pre-alarm system can avoid the occurrence of major social events.
To establish this pre-alarm system, we first detected the social events from the complicated online data utilizing the modified Dynamic Query Expansion (DQE) method and emotion analysis technology. Then to achieve timely tracking and prediction of the event participants’ social attitudes, machine learning methods were used to establish the social attitude predictive models based on the users’ data on Weibo. Finally, we built the pre-alarm system based on the social attitudes, emotion and scale of the event participants. This pre-alarm system was built from the real data of previous social events and can provide early judgement of whether the social event will develop in to a major event.
The current study was an interdisciplinary research that incorporated knowledge of psychology, computer science and social sociology. We detected 88 meaningful social events with the DQE method and emotion analysis technology, with a precision of 0.78 and a recall of 0.74. The predictive models of social attitudes have a moderate pearson correlation that ranged from 0.47 to 0.62. The pre-alarm system that predicts the trends of social events reached the precision 0.78 and the recall 0.88.
This paper proposes to predict the trends of social events on social media. Based on social psychology and communication research, we extracted unprecedented amount of comprehensive and effective features which are relevant to the tendency of social events. Then we built predicting models utilizing regression algorithms in machine learning.
The results of these experiments demonstrated the effectiveness of our proposed approaches.