基于社交网络的群体性事件挖掘和预测

Unrest Event Mining and Prediction Based on Social Network

文本事件挖掘旨在通过挖掘文本来实现结构化的事件表示,从而支撑进一步的事件分析和预测工作。文本事件挖掘所需要挖掘的结构化事件信息包括事件类型、参与者、触发词、时间、地点等,其中每一项信息的挖掘都是一个单独的文本分类或者是信息抽取任务。所以,文本事件挖掘是一项综合性的自然语言处理任务,具有较高的工程复杂性。社交网络群体性事件挖掘是指针对社交网络这一特定的信息源,以及群体性事件这一特定的事件类型所开展的文本事件挖掘工作。由于社交网络和群体性事件的特殊性,发现针对社交网络群体性事件的挖掘结果可以作为事件预测的直接线索,因此,实现了一个基于社交网络的群体性事件挖掘系统。在该系统中,实现了对文本信息事件发现和分类、参与者抽取、行为抽取、地点抽取和时间抽取等子任务,共同组成完整的事件要素结构。同时,根据事件时间信息进行事件预测,并与实际事件发生情况进行对比以评测事件生成效果和预测准确率。

Text event mining aims to realize structured event representation through text mining,so as to support further event analysis and prediction.The structured event information to be mined for text event mining includes event type,participant,trigger word,time and place,etc.,among which each information mining is a separate text classification or information extraction task.Therefore,text event mining is a comprehensive natural language processing task with high engineering complexity.Social network group incident mining is a textual event mining work carried out by pointer to the specific information source of social network and the specific event type of group incident.Due to the particularity of social networks and group incidents,we found that the mining results of social network group incidents can be used as the direct clues of event prediction.Therefore,we have implemented a group incident mining system based on social networks.In this system,we realize sub-tasks such as text information event discovery and classification,participant extraction,behavior extraction,location extraction and time extraction,which together constitute a complete event element structure.At the same time,we make event prediction based on the event time information and compare with the actual event occurrence to evaluate the event generation effect and prediction accuracy.