Social Network Bursty Topic Ouary,Prediction And Visualization Based On Spatiotemporal Characteristics

With the booming development of the Internet,social network has gradually occupied a vital part of people's lives.Nowadays the social network platform represented by Weibo gathers all kinds of information in the society.Users reveal events,spread news and share experiences on social networks.Social networks represented by microblogs have influenced everyone's daily life.Users have generated a series of behaviors such as sharing life experiences,interests and hobbies,discussing street current affairs and hot public opinions on social networks.The bursty topic query and visualization system based on the technologies of bursts detection and tracking plays an important role in the task of network public opinion analysis.It can improve the efficiency and quality of data acquired by the public opinion analysis system,help relevant departments and enterprises quickly understand the development pattern of bursts,and provide accurate information support for timely decision-making of public opinion guidance.The main work of this thesis includes the following aspects:(1)Based on spatiotemporal characteristics,this thesis proposes a bursty topic detection method(Burst_NBT)to detect the bursts in Weibo,a typical social network platform.Burst NBT uses spatiotemporal characteristics to detect the bursty words in the corpus and filters the noise in the data.Then it carries out agglomerated hierarchical clustering on the bursty word set,calculates the bursty score and sorts the candidate bursty clusters.Lastly,it detects Top K bursty topics efficiently and accurately.The experimental results show that Burst_NBT algorithm can effectively detect the bursting topics in the microblog data stream,and is superior to the comparison algorithm in terms of the accuracy of topic detection and the comprehensibility of topic expression.(2)This thesis proposes a social networks bursts tracking algorithm based on semantic extension(SADV-SE).SADV-SE algorithm uses the wiki knowledge base to extend the semantic features of the posts after feature selection.Then,the posts are transformed from the weighted feature set to vector representation,and the posts to be tracked are classified by the similarity between the topic vector and the posts' vector.During this period,the adaptive learning of classification threshold and topic model updating are carried out,and the classification processing of the posts is continued until there are no posts to be tracked.Experimental results show that SADV-SE is superior to the contrast algorithm in terms of precision,recall and F1 value.(3)This thesis proposes a framework for querying bursty topics and an attention prediction algorithm of social network bursts based on neural network(MA-LSTM).By analyzing the factors that affect the attention of bursting topics,this thesis utilizes topic content characteristics,topic social attribute characteristics and topic participation user characteristics to predict topic attention.The parameters are learned through neural network modeling based on two-layer LSTM.The experimental results show that the MA-LSTM algorithm can effectively predict the attention of bursty topics on Weibo,and the experimental results are better than the comparison algorithm in both the residual error and the relative error.(4)This thesis designs and implements a query and visualization system for bursty topic based on spatiotemporal characteristics of social networks.The system consists of four modules:bursts detection,bursty topic tracking,attention prediction,query and visualization.Bursty topic detection and tracking identifies bursts and their subsequent microblogs in the data stream.The attention prediction module realizes the prediction of the follow-up attention of bursts.The query and visualization module provides interface for query and visualization of bursts and their microblogs.The system has certain practical values,and effectively verifies a series of algorithms proposed in this thesis.