related.tex

\section{Related Work}
\label{sec:related}
\noindent\textbf{Characterizing the Scams in Blockchain Ecosystem.}
A number of studies have been focused on scams in blockchain ecosystem. Some of them focused on the code level scams in platforms such as Ethereum and EOSIO. Torres et al.\cite{torres2019art} demystified  honeypots in ethereum smart contracts, Bian et al proposed a deep learning system to detect the ICO projects\cite{bian2018icorating}. Xia et al. \cite{xia2020characterizing}characterized scams in cryptocurrency exchanges. In analysing Bitcoin, Paquet-Clouston et al.\cite{paquet2019spams}characterized sextortion in the bitcoin ecosystem and uncovered scammers' operations. Vasek et al.\cite{vasek2018analyzing} characterized Ponzi scheme in bitcoin. The above researches focus on some types of scam, while there is no integrated analysis of all types of scams.


\noindent\textbf{Bitcoin address clustering and analysis.}
When dealing with scam Bitcoin addresses, these address always work in cluster in reality, by analysing the address clusters we can find the illicit campaign behind the scam behaviors. Xia et al.\cite{xia2020characterizing}identified and characterized the cryptocurrency exchange scams, They found 94 scam domain families and 30 fake app families. Some research proposed new ways of clustering bitcoin\cite{ermilov2017automatic, zhang2020heuristic}

\noindent\textbf{Implementing graph neural network on Bitcoin.}The transaction mode and ledger in bitcoin makes it more comprehensive to analyse bitcoin in graph methods.  Fleder et al uses the graph-analysis framework to find and summarize activity of known and unknown users\cite{fleder2015bitcoin}. Most of the transaction graph related researches are related to the discussion of the anonymity in Bitcoin. Haslhofer et al presented GraphSense, an cryptoasset analytic platform, which can be used for interactive investigations of money flows, and more advanced features\cite{haslhofer2021graphsense}.

Weber et al, released Elliptic dataset, and performed classification using several benchmark methods and GCN\cite{weber2019anti}, however this dataset is an anonymous one. The features are all anonymous. Besides the topological feature of this graph is not very complete as the unknown addresses do not come from the transaction relation with the labelled addresses.