Research On Community Detection Based On Bitcoin Anonymous Transactions

With the widespread use of digital currency,the illegal trading activities of bitcoin have been paid attention by regulators.However,bitcoin has a large amount of data and weak information relevance.How to mine user entities from encrypted address transactions and associate illegal transactions with users has become a hot issue in the current research on de-anonymization.In view of the above problems,this thesis aims to mine the community attributes of bitcoin transactions,and conducts the following research in network construction division and community attribute classification prediction.(1)Construction and division of bitcoin user network.Heuristic rules are designed based on the transaction mechanism,and the real-time bitcoin transaction network is constructed by integrating the bitcoin dollar value data.On this basis,a Louvain based algorithm for discovering bitcoin hierarchical overlapping communities is proposed,which can effectively identify overlapping communities in user networks while retaining the ability to quickly divide hierarchical communities.(2)Aiming at the problem that the attribute of bitcoin network community cannot be recognized,a classification model of bitcoin community based on RF-Softmax is proposed.On the basis of community division and core address label data set,this thesis compares the five common label attributes of the special currency community from the perspective of community core entity characteristics,community characteristics and the joint characteristics of the two,so as to achieve the effective classification of the dark net and the mining pool community.(3)In view of the limitations of mining features from community central entities and community features,Graph Convolutional Network based Bitcoin Community Classification Model(GCNBC)is proposed to further improve the classification accuracy of complex communities.The multi-layer pool convolution operation is integrated to mine the deep characteristics of community users and transactions.Design a feature sort pooling to extract important influence features under different community structures and output them as a fixed number of features.Local feature extraction is introduced to achieve better end-to-end multi classification effect in complex trading communities.The results show that it is feasible to excavate the illegal activities of bitcoin from the perspective of community.Identifying the community type where the collective user behavior exchange is located helps to timely find abnormal communities and locate abnormal user trading activities.