to leave a comment.
University research team under China's Ministry of Public Security: "Responding to illegal cryptocurrency transactions and economic crimes"
A research team at China's police academy has developed an artificial intelligence (AI) system capable of detecting illegal cryptocurrency transactions such as Bitcoin money laundering, the South China Morning Post (SCMP) reported on the 2nd.
According to the report, a research team at the People's Public Security University of China (PPSUC), which is equivalent to China's police academy, has developed an AI framework capable of detecting illegal cryptocurrency transactions with 90% accuracy.
Dr. Sun Jingchao, the corresponding author, explained in a paper published in May in the Chinese academic journal 'Journal of Intelligence' that these research findings provide an accurate, generalizable, and interpretable solution for detecting illegal cryptocurrency transactions.
He, who specializes in criminal investigation and cybersecurity research, added that these research findings will be an innovative way for regulatory authorities to respond to illegal cryptocurrency transactions and economic crimes.
SCMP pointed out that this research was conducted amid strengthened crackdowns by Chinese authorities on financial crimes related to virtual currencies.
In March this year, the Supreme People's Procuratorate announced that it had prosecuted 3,259 people last year on charges of money laundering using virtual currencies and underground banks.
Although Bitcoin transaction details are recorded on a public blockchain, wallet addresses are not directly linked to the identities of individuals or institutions in the real world. Illegal funds can also move through multiple complexly intertwined wallet addresses.
Detecting such continuously evolving transaction patterns with traditional systems alone is becoming increasingly difficult.
Dr. Sun stated, "Our method has significantly enhanced its generalization capability to identify newly emerging or evolving illegal transaction methods."
Generalization capability refers to the ability of an AI model to perform well on new data outside of its training data.
The research team explained that to analyze cryptocurrency transactions, they combined Dynamic Graph Neural Networks (DGNN), memory mechanisms, and Large Language Models (LLMs).
DGNN assumes the cryptocurrency ecosystem as a constantly changing network and analyzes transaction structures, fund flows, and changes over time.
The memory module finds and compares new transaction cases with similar past illegal transaction patterns.
The LLM converts these complex results into natural language risk assessments. It not only presents whether a transaction is illegal but also provides a risk score, key evidence, and the reasoning process.
The model achieved an overall accuracy of 89.4%.
The precision for illegal transaction detection was 89.1%, and the recall was 64.5%.
'89.1% precision' means that 9 out of 10 transactions classified as illegal by the AI were actually illegal. '64.5% recall' means that the model identified approximately two-thirds of the transactions that were actually illegal.
The research team stated, "Tests using a public Bitcoin transaction dataset proved that it performs better than major baseline models."
Newsletter
Get key news delivered to your email every morning
to leave a comment.