Social Event Forecasting Based On Precursor Sequence Mining

Social events refer to group activities such as sports games,academic activities and political activities,which are organized,have specific participants and gather at a certain time and place.The occurrence of such incidents often has a significant or even negative impact on social stability.For example,the Floyd incident in Minneapolis,Minnesota in 2020 led to demonstrations in more than 30 cities in the United States,which had a great negative impact on local politics and economy.However,because the social system is extremely complex,social events involve a large number of groups,organizations and individuals,and the incentives are diverse,so it is not easy to predict them.Fortunately,the advent of the era of big data has brought a new opportunity for this research: the popularity of mobile Internet and social media makes it possible for Internet users to obtain and share information at any time.The Internet is becoming a global "social sensor network".From the open-source big data such as news and social media,we can dig out the inducements or precursor signals of social events,we can make perceptual prediction.This paper focuses on the prediction technology of social events,mining the inducement or precursor events from the open-source big data,and making full use of the feature mining ability of deep neural network,constructs the prediction model of social events based on the incubation process,so as to provide more accurate decision-making reference for the government and policy makers to prescribe medicine for the disease and prevent it in the future.This paper mainly carries out the following three aspects of research work.First,data set construction.For social event prediction task,the quality of data set determines the upper limit of prediction model.In order to realize symptom mining and training and evaluation of prediction model,we need to build a data set containing the target event and its breeding process.In this paper,we choose GDELT(global data on event,location and tone)and ICEWS(integrated conflict early warning system)event data to construct data sets,in order to find the target event and its breeding chain in open-source data,reasonably define the relationship between events,and associate them into a chain according to the time of event occurrence.Second,the hierarchical symptom mining.Detecting precursors from continuously acquired event data is the fundamental premise to accurately predict whether social events will break out.In order to quickly identify and extract precursors from the event breeding process,firstly,the event precursors are defined,and then the hierarchical symptom mining algorithm based on EF-ICF is used to mine typical precursors from the target event breeding process,that is,to identify typical precursors,form a hierarchical classification system,and establish a symptom knowledge base.Third,the prediction of social events based on symptom sequence.In this paper,event prediction task is defined as a binary task.This chapter introduces the latest research ideas related to the most relevant tasks of this paper,points out the shortcomings of this kind of research,and proposes a prediction framework based on the sequence of incubation process graph,which makes full use of the mining ability of graph neural network.The experimental results show that the proposed model has a good prediction effect on multiple data sources.At the same time,taking the typical symptom sequence as the input,the representation learning of different levels of precursors is carried out to form the embedded representation of the similarity of symptom chain.The experimental results show that the introduction of symptom sequence improves the prediction effect of the model.To sum up,this paper focuses on the research of social event prediction technology for open-source data,and proposes a hierarchical mining algorithm based on EF-ICF,which can identify typical precursors from the event breeding process,and establish a hierarchical system to provide support for online prediction;Using the good mining ability of graph neural network,this paper focuses on the event prediction method based on incubation process graph sequence.The experimental results show that the prediction model proposed in this paper has good prediction effect.