Research And Analysis Of Social Network Event Detection Methods Based On Node Evolution

Event detection is an important means for the government and enterprises to master sensitive topics and public opinions,which can make the network society more harmonious and progressive.The event detection methods of traditional media and social media are only suitable for the standard written news reports and the sparse,dynamic and social short texts,respectively.Both methods require text processing,which is tedious and inefficient.The event detection method based on network evolution has the characteristics of both simplicity and efficiency.However,the directly established network evolution model only analyzes the network evolution at a macro level and ignores the evolution of nodes.The network evolution can be accurately analyzed by considering the network evolution with nodes.Therefore,we propose an effective method for event detection on social networks using link prediction technology from the perspective of node evolution.An event detection method based on node evolution staged optimization(NESO?ED)is proposed to solve the low sensitivity and poor stability problems,which are caused by the method based on the best link prediction index.The nodes in the network are not always following the same evolution mechanism in the evolution.The network evolution is analyzed by the best link prediction index instead of the fixed index.When the best link prediction index of nodes is selected,the network evolution is divided into stages.Then different optimization methods are adopted to quantify the network evolution according to the characteristics of different stages.The experimental results based on three data sets,i.e.VAST,Dept1 and Dept2,demonstrated overall proportion of the best indexes of nodes during the network evolution process and the changes of these indexes in each stage.The results indicated that this method had high sensitivity,and the stability of event detection was also improved.It is suitable for situations which the occurrence of an event has the smaller effect on the network structure.The event detection method based on coarse-grained node evolution network(NENC?ED)is proposed to improve the stability of the method based on the fixed index,analyze the abnormal evolution of nodes,and deal with the complex network structures.When the network remains stable,the nodes tend to keep a fixed evolution mechanism.But when the network fluctuates,the nodes will change evolution mechanism.So the abnormal nodes play an important role in the network evolution.In our thesis,the first step is to extract nodes with abnormal behavior during the network evolution process.Then the increasing amount of user data makes the network scale larger and larger,it results in a large amount of algorithm calculations and low event detection sensitivity.So the coarse-grained network is applied to simplify the network structure in the second step.At last,the abnormal evolution nodes are introduced into the coarse-grained network for event detection.Based on the experiment results on VAST,Dept1 and Dept2 data sets,the changes in the number of abnormal node evolutions during the network evolution process were analyzed.The experimental results showed that the method had strong stability,and the sensitivity of event detection was also improved.It is suitable for situations which the occurrence of an event has the larger effect on the network structure.