In　this　paper,　we　investigate　the　challenge　of　cryptocurrency　transaction　tracking　through　dynamic　network　link　prediction.　Firstly,　a　transaction　tracking　model　for　Ethereum　cryptocurrencies　is　proposed　(DNLP-TCT),　which　employs　a　hybrid　approach　involving　temporal　attention　mechanisms,　Bi-directional　Long　Short-Term　Memory　Network　(Bi-LSTM),　and　Autoencoder.　Then,　by　leveraging　the　power　of　these　machine　learning　techniques,　the　model　implicitly　captures　the　dynamic　behavior　and　structure　characteristic　of　the　transaction　network.　This　enables　the　prediction　of　future　links,　facilitating　the　inference　of　potential　future　transactions.　Finally,　the　model　is　evaluated　on　three　real　Ethereum　transaction　datasets,　and　performance　is　compared　against　several　representative　methods.　Based　on　the　results,　the　proposed　model　exhibits　significant　improvements　in　both　Area　Under　the　ROC　Curve　(AUC)　and　Error　Rate　metrics,　validating　its　effectiveness　in　the　field　of　transaction　tracking.　　©　2024　Asian　Control　Association.