Research And Implementation Of Bitcoin Address Relationship Analysis Algorithm Oriented To Transaction Dimension

Bitcoin is known for its liberty and protection of privacy.Its transaction data is public,and its privacy protection relies heavily on the anonymity of addresses.In the study of Bitcoin privacy,Bitcoin address relationship analysis is crucial.Today,the development of Bitcoin privacy-enhancing technologies,including coin mixing algorithms,is making traditional address relationship analysis algorithms difficult to use.This thesis focuses on the Bitcoin address relationship analysis based on transaction data under the presence of coin mixing,first analyzes the Bitcoin data structure and then implements the Bitcoin on-chain data simplification method,the Bitcoin data tracing algorithm,and visualization algorithm for trans-action and address dimensions.The proposed transaction tracing method can convert the traditional method’s directed acyclic graph into a tree.Finally,this thesis proposes the LSTM Transaction tree Classifier(LSTM-TC)which is a method used to detect coin mixing transactions and an address relationship de-termination model using LSTM.LSTM-TC helps the traditional heuristic rule-based address clustering al-gorithm to identify coin mixing transactions,which can get better address re-lationship results.Although currently coin mixing detection algorithms are widely used in practical address clustering methods,the algorithms used are all based on manually designed rules that are difficult to adapt to new coin mixing classes,but LSTM-TC,which is based on deep learning,solves the problem to some extent.In the study about this approach,a coin mixing transaction dataset was built using the well-established coin mixing detection algorithm,which is used for training and validating the coin mixing detection model.Then,based on the characteristics of Bitcoin transaction data structure,we proposed a Bit-coin transaction tree feature extraction algorithm.Sequence features of the Bit-coin transaction tree were fed into the LSTM classification model to obtain the result.The model is trained and evaluated on the dataset and compared with other methods.The experimental results show that the supervised learning-based coin mixing detection method proposed in this thesis has a high recall ratio and helps to obtain more accurate clustering results.In the address relationship determination model using LSTM,all trans-actions associated with one address will be considered together.Bitcoin ad-dress features will be extracted through a manually designed feature extractor,and then the relationship between two addresses will be determined by a deep learning model.While traditional address clustering methods derive address relationships based on only one transaction at a time,this method derives the features of the address based on all associated transactions of it,which can theo-retically obtain more information for addresses with a larger number of transac-tions.This thesis also creates an address relationship dataset using early data of Bitcoin for model training and model effectiveness evaluation.Then an LSTM-based Bitcoin address relationship judgment model is proposed.Experiments show that the address relationship model has a certain accuracy rate and can be used to determine a small number of address relationships quickly.The bitcoin address relationship analysis method proposed in this thesis can be used to implement address de-anonymization related tasks such as bit-coin address clustering,which further helps to applications such as bitcoin fund tracing and transaction classification,and can be used in scenarios such as finan-cial supervision and illegal fund investigation.In turn,the approach proposed in this thesis provides some inspiration for the improvement of Bitcoin privacy enhancement techniques,especially coin mixing technology.