Tao Yan - Multi-Layered NFT Wash Trading Detection Leveraging Sequential Transaction Features

Transcript

hello everyone welcome to my presentation I'm toan I'm from the usage blockchain Center and the blockchain distributed nature group uh today I'm going to present our This research called multi- layered nft or trading detection leveraging sequential transaction features and this work is collaborated with Dr Shan and also my professor Claudio uh here is of this presentation uh let's first go to the have a brief introduction of the nfts so nfts are the long functionable tokens and uh it uh all these nfts are unique

with which means they are not uh replaceable and interchangeable so there are three nfts corrected and inut different colors they are they are L equal but in our this work we are more focused on the unique ownership of this uh nft so which means uh at a specific time this each nft just belongs to one unique user this is very important in this research to detect the O the O trades uh so what is o trading it is comes from the transitional fance uh Ur o trading is a market

manipulation uh strategy where an investor can repeatedly sells and buys the same asset for multiple times and this activity will lead to the misleading Market activities so they will create artificially create more uh invented the tradings like to deceive other users to to believe yes this nfts are popular but actually there are lot uh so this blockchain and this unique ownership of Mt provide us a opportunity to detect this nft or

tradings why you see in the N colon there is a transactions on the blockchain and uh in the mid colon is this is a Marketplace this Marketplace is also unchain like the open SE uh they have this uh n f are traded here and they record the price of nfts and so we combine this transactions and also the price information and then we use the network to re to represent the uh tradings and the transactions of the nfts but here I think is notable that uh for these transfers we are

talking about there are lot just about the tradings it can be also the transfers between uh two uh addresses which there is no there is no price so like I sent the nft as a gift to my friend and there is no price information okay uh so here we focused on each the uh the history transactions of each nft and we represent the transaction history as Network for each nft uh here I want to talk about more about this sequential transaction

feature what it it exactly means here is a table which shows there are two transactions of this uh B ayc nft and uh their token ID is one20 but uh they have the different the addresses of the cender and the and the recipient uh what means this sequential transaction is that the cender of the next nft transaction is always the recipient of the previous transaction so this Con construct a chain of ownership change I think this is very you this is very different uh for nft comparing with

erc20 tokens because erc20 tokens are funable tokens and you don't know like the erc20 you hold and you send it to other people and other people will also get the same funable tokens from from the third people so but nft will not have this uh problem every nft just belongs to a unique user at a specific time here comes to the research questions so our main research question is that how to effectively detect the O tradings like for for nft in the whole

life cycle of of the nft and we are also interested in like if there are any Traders they are involved in different multiple or multiple nfts for the watch trading activities uh we connect the data from the AR cavalot which we are running in our uh group and uh in this research we focused on the B app yach car it's called shortly C is BC which is very popular and has the highest trading volume among all the nfts uh we here we connect all these transfer events from the start of this

nft to uh May 4 of 2023 and we also collected the price information from multiple nft Market places this including openc the berar and also the nuks r and x2y2 in total we have got around around 200 sound transactions and 40 and 34 sound prices informations yeah here as I said before this PR is not every transaction where has this price information here comes to the m Dage part so how can we finally detect the

nft was trades the basic idea is still we Lage the sequential transaction feature of this nfts and it can it it contains three steps the first step is we uh convert these transactions into into aray which just include the sender and the receiver address but they are in the uh time order and the next step is that we start from the uh first address and the we detect the cyclonic structure so the cyclonic structure is

that from the AR we check which are the first place which are the first addresses are repeated so like for a b c a and yeah we found a are the first uh address is repeated if we have a b a b c b yeah and then B is the first address repeated and we also check after this repeated address if that address is also occurs before yeah if it occurs before we also included that address uh into the first psychical structure after we detect the first psychical structure we have the

first uh uh SP spaces nft watch reading and then we want to see if this or trading continues happen in this nft so we detect the multi ners so we're following the similar logic DET detec in the first layer but this step increase another additional uh state is to check whether this address has already appeared in the previous ler we have some result to using this methods we have generate some results I

think the first results is to generally show the Dynamics of the nft this nft connection trading histories so here we visualized the number of nfts traded uh by by each day and also the number of transactions and also the trading volumes in each and also how many Traders for each day to they are trading this nft what is interesting from this PL is that the number of Traders involved is always much lower than the number of trading the transactions so this is uh so ideally for because each transaction you have you will have

two uh Traders but here we found the trader the number of Traders is smaller than the number of the transactions so what happened there uh yeah if we look at this plot it will be more easier to understand so ideally because each transaction will have two Traders the number of Traders will go linearly increase with this number of the transactions but here in actual you see there are two blue line and the yellow line but the number of Traders are smaller than the number of transactions so what happened here is that in these nft transactions some

Traders repeatedly buy and sell these nfts and uh yeah it uh needs to the total number of these Traders uh reduced and also here it indicates there are Cycles inside of these transaction networks here we visualize that there are two nfts transaction networks we do find that there are some Cycles inside of this uh single nft transaction networks so in uh here we want to uh give our def in our work of about what is nft watch trading in the context of Network watch

trading means the circles with price increasing uh here we have a we have a simple Network to show the o trading happened in one of the nfts in this uh connection the nft is 4200 and it's it's if you look at the plot it starts from the uh point in Orange and uh goes through the cycle uh you see the price which is the the small number of the age and it continues to increase and finally it

increased more than 30 times for this uh nft uh but here here uh we are more interested in like the H uh the H records of this or trading activities in one in a single nft so we detect the multi layer uh or tradings so here what means ner the N is means um so we so first thing we start from the psychical transaction which is in the newest uh uh in Green in in the blue color yeah this is a cical uh

structure and this is a first n and so when is this first first n stopped when you found there is a new address which didn't appear before so it it's it's the end of the first uh uh nature and then it continues to grow when they have the next find the next psychical structure and it also finds there is a new Trader and it start and it stops again this is the second new year so it continues to grow and finally with for this nft we found that there are four layers and all layers demonstrate the

dynamic was traing activities through the KN SP of a single nft uh here we also have some uh results for the features of this nft was trading activities the left pel shows there these uh was trading activities are centralized in the newer ners so if the ners is smaller yeah they have more thiss are are more happened are more likely happened on these layers but if you wor you you can have this um like around 80 layers for a single nft the

right panel shows some uh time features of the nft transaction of the watch tring so basically you can find the average length of the duration of the initial uh initial layer is 42 days yeah and then we want to find if there are some Traders like they are involved in multiple nftd or trading activities so we use a key core in the network science to identify so who are the in the research in the core groups for each nft network uh here we start from the Three

core uh structure and then we continue as a four core structure but yeah here we find some the addresses which is showing in the XX uh you can see uh the first address which ends at 2762 it has it involved uh 338 nfts in this connection and we also have other uh addresses so here we just show the top 10 addresses who involve most of this uh nft or

tradings uh yeah let's make a brief conclusion of this work so we Leverage The sequential trans transaction to detect the uh wash tradings and we found there are multi- layered uh was trading parns in this uh bayc nft connection and to use this method we can avoid to have abundant these Cycles or tered because the previous resarch research they they use the Deep first search or the brace first search to using these methods they will have abundant of Cycles but in this in our

work it just will return the whole life knif span of the nft transactions of a single uh nft uh finally there are some take home messages out of this 10,000 NS around five uh s anties involved in o trading and among these o tra anties around 3 S enties have multier aary importance and finally we identified there are 80 83 Traders they participate in multiple nft uh watch treatings yeah that's for former presentation thanks

for your [Applause] attention thank you so much do we have any questions from the audience I okay thank you thank you for the presentation so my question is uh about the what happened next like you identified uh 180 wor Traders right did you do any further action like report them to somewhere and like some kind of institution like police or whatever sorry I I don't know if I understand your question correctly you mean do we report these twers yeah L what happens next like you you

identified them and what's what now uh yes we I think it's very interesting to say who are doing this uh frequently doing these activities but uh currently we are doing this research work so we will publish this work like in some conference or joural but yeah we current we didn't think where where we can public this Transit I think finally we can generate this data set yeah and this data set can be public yeah uh can you just please wait for the mic yeah there is a volunteer coming so you just checked for direct transfers of

nfts and without any transfer of ethereum or anything back uh sorry can you repeat the question so for example I was just looking lately in uh all the kind of nft was trading and I saw there's a group they they bought prate APS they send it to another address and then they get exact amount of ethereum into their wallet B yeah did you include actions like this or not you mean they traded on different marketplaces no direct transfer and direct transfer of etherum in not over the marketplace you mean self

transferred yeah yeah yeah we have we have include this so if person can the same person they bite and they also can be can send this nft to himself but uh in this case this is just a self Loop and we have this in our we can detect this in our data set but we for the results we ex excluded that yeah because in this part they didn't uh like to involved other in other uh Traders and users yeah any other oh yeah there is one more

question right there hi um I'm just curious if there are any implications from the data that um for example traders that involve in multi-layered um was trading B profit more uh than the traders that involve only in one layer of was trading is there any implications from your data okay yeah this is very interesting uh Point currently we didn't check if they will preit more to have this multiple layers but we do uh by the way yeah for sure we can we can do that because we have the data for each Trader and how many price they have increased by this

uh activities yeah I think uh later I can yeah I will generate the real thank you thank you all any other questions now then please give another big round Applause to to thank you so much for your talk thanks [Applause]