Research And Design Of Bitcoin Transaction Behavior Analysis System

Users on bitcoin are highly anonymous,which also breeds many illegal behaviors through bitcoin,such as dark network,money laundering and so on.Considering that these illegal behaviors will reflect some characteristics in bitcoin trading behavior,this thesis analyzes the data on bitcoin from the perspective of trading behavior,so as to achieve a certain degree of de anonymization,which can provide help for regulatory agencies and law enforcement departments in detecting and obtaining evidence for illegal transactions on bitcoin.Traditional methods for analyzing transaction behavior on bitcoin mainly combine entities through heuristic methods,analyze their commonness through entities,or predict through feature engineering,and then use machine learning models to predict.However,these methods ignore the connection characteristics between bitcoin data.Bitcoin data can be constructed into directed acyclic graph,in graph,structural characteristics feature is difficult to describe in traditional machine learning.In this thesis,we use graph neural network to process the bitcoin data,and the graph structure features on the bitcoin are effectively used,so as to improve the performance of the model on the bitcoin data.This thesis proposes two methods to analyze transaction behavior on bitcoin:(1)Bitcoin transaction behavior analysis method based on time sequence information,the transactions on the bitcoin have sequence,which can be described by dividing the bitcoin data into time sequence intervals,and then the graph embedding is obtained by GAT(Graph Attention Networks)in time sequence intervals,and the time sequence information is obtained by GRU(Gated Recurrent Unit)between time sequence intervals.(2)MLP(Multi-Layer Perceptron)combined with graph neural network bitcoin transaction behavior analysis method,because the data on the bitcoin has more complex characteristics,graph neural network will aggregate the characteristics of neighbor nodes to the target node in the training process,which may submerge the characteristic information of the target node itself.Therefore,this thesis retains the characteristics of the node itself through MLP,and aggregates the neighbor node information through GeniePath,This method can achieve very good results in the case of complex data characteristics.This thesis also designs and implements a bitcoin transaction behavior analysis system,which can combine the traditional method and graph neural network method to analyze transaction behavior of bitcoin,and de anonymize the source and whereabouts of its funds to a certain extent,so as to provide reference for regulatory agencies and law enforcement departments to detect illegal behavior and obtain evidence.