Pay attention to Bitcoin: utilizing attention mechanism with lightning network information for change address detection

Abstract

As Bitcoin’s popularity has surged, its scalability has been challenged by limited block capacity, leading to increasing transaction fees and diminishing its suitability for microtransactions. The Lightning Network (LN), introduced by Poon and Dryja as a second-layer solution, offers near-instantaneous transactions with minimal fees, making it an attractive alternative for Bitcoin users. However, as LN adoption grows—securing over $336 million in Bitcoin—concerns arise about the potential compromise of user anonymity due to the transparency of public LN channels. Existing research has primarily focused on privacy within either the LN or Bitcoin networks independently, leaving the cross-layer privacy implications underexplored. This study investigates the impact of LN activity on Bitcoin user anonymity, specifically focusing on the detection of change outputs in the Bitcoin network. We analyze the transactions and their inputs and outputs, comparing standard Bitcoin transactions with those linked to LN channel creation. Leveraging a modified Bidirectional Encoder Representations from Transformers (BERT) model, we enhance our ability to identify change outputs, achieving a 93.9% success rate—an improvement over the baseline. Our findings reveal significant privacy risks posed by cross-layer interactions between the LN and Bitcoin networks, demonstrating that data from the LN can be effectively used to deanonymize Bitcoin transactions. This research provides critical insights into the privacy implications of the Lightning Network and lays the groundwork for further exploration of cross-layer deanonymization risks in blockchain systems.

Date	3 Nov 2024
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Kaiwen Zhang (Supervisor)