Research On Social Media Event Detection And Evolution Methods

Social media is a content production and exchange platform based on user relationships on the Internet.The rapid expansion of social media in recent years has allowed people to easily understand what is happening in the world in real time.Event detection based on social media is a technology that digs out valuable information from massive social media content.Based on the events detected in social media,the national management agency can timely understand the major emergencies in the real society and take corresponding measures.Individuals can learn about the hot topics that appear in the society in time and participate in discussions.Among social media event detection methods,online event detection has attracted the most attention of researchers due to its high timeliness.Online event detection methods usually first use text online clustering technology to group social texts,then extract event features for text clusters,and finally set thresholds based on feature values??or use supervised classification models to filter out events.However,the existing methods are as follows: There are shortcomings: first,due to the sparse and noisy social text information,the commonly used online text clustering methods have the shortcomings of cluster center drift and unstable clustering effect;second,the text cluster event characteristics obtained by online clustering It is not obvious,which makes it difficult to determine the event.At the same time,the uncertainty of the value range of the event characteristics of social media data from different sources makes the event determination model not universal.In response to these problems,this article takes the Twitter data stream as the research object and real-time online event detection and evolution as the goal.The main innovations are as follows:(1)Propose a social text online clustering method based on word burst features.Aiming at the problem of cluster center drift and unstable clustering effect in the current online text clustering method used for event detection,this method uses the emergent features of the event to cluster the event text as the center,and first detects the emergent words,And then use the spectral clustering method to divide the recognized sudden words,and use each division result as a cluster center,and then cluster the social text with a fixed cluster center.The clustering method can achieve the effects of fixed cluster centers and stable clustering results during real-time clustering.(2)Propose an bursty event judgment method based on the prototype network.This method first extracts temporal features,text consistency features,and user diversity features based on the suddenness and evolution of events for text clusters.This feature extraction method can extract event features of varying degrees and durations,and solves the event features of text clusters.Inaccuracy leads to the problem of difficult event determination.Aiming at the disadvantage that the existing event judgment model is not universal,this method adopts an adaptive semi-supervised method to determine the event on the extracted text cluster features,and uses the prototype network method in small sample learning to measure and learn the event features,so that the judgment model can be When changing the detection data source,only a small number of samples need to be marked to obtain a good event determination effect.