Research On Recognition Of Social Media Event Attribute

Social media platforms have a large number of users and are full of information.Whenever there are major events in the real world,the relevant information of these events will be transmitted to social media,so mining the attribute information of social media events has great value.Recognition of social media event attribute is to analyze the social network text data to extract information such as the time of the event,the people involved,the geographical location of the event,and the topic to which the event belongs,so that it can obtain richer information about the event.By identifying the event's attribute information,public opinion monitoring and analysis can be realized,and it is convenient for public decision-making agencies to obtain information and handle it in time.Event attribute recognition involves a lot of content.This article mainly focuses on identifying the topic of the event and where the event occurred.In the existing research,methods for topic recognition rely on the external characteristics of social media,so they are not portable across different social media platforms,and these methods cannot be applied to online topic classification.Most of the methods for geographic location recognition rely on a single geographic information for extraction,which leads to insufficient information utilization.At the same time,the existing methods do not effectively deal with the noise in geographic information,resulting in low accuracy of geographic location extraction.This article conducts research on the above issues.The main contributions and innovations are summarized as follows:(1)An event topic classification method based on topic words is proposed.This method analyzes the time-varying characteristics of events and words in the social media information stream on the topic distribution.In the topic word extraction process,the sliding window is used to calculate and update the topic distribution of the words in real time,and the KL divergence is used to calculate the difference between the event topic distribution and the word topic distribution,and the KL divergence value of the words is used to extract the event topic words.In the topic classification process,the topic words contained in the event are extracted,and the probability distribution of event topics is calculated and updated in real time by using Bayesian inference method based on Dirichlet-polynomial conjugate distribution.This method does not use the external features of social media,and calculates the event topic probability distribution in real time,so it is portable and can implement online topic classification.(2)An event geographic location extraction method based on multi-source information fusion is proposed.This method analyzes the geographic attribute characteristics of tweet text,Twitter context,and user personal information,and proposes a method for constructing a basic probability assignment using the above three geographic attribute information.K-Means clustering algorithm is used to remove noise points from user personal information sources,and the influence of the population density distribution on the basic probability assignment is eliminated through normalization.In the process of location prediction,the basic probability assignment is fused by using Dempster synthesis formula,and the probability interval of each location is calculated by using the reliability function and the plausibility function.Finally,the geographical location of the event is judged by the probability interval.Compared with the existing event localization methods,it is proved that our method has higher prediction accuracy.