Mapping Ethereum with AI

Find everything from NFT hodlers to institutional wallets

The Map
8 min readSep 21, 2022

Imagine a world without maps. We can learn about a place by saving an address from conversation or the newspaper, then search that address in a database. We’d have no way to view neighborhoods, discover locations of interest, or navigate. With on-chain activity, discovering participants, contracts, and communities is also anecdotal. Interesting accounts, projects, and communities are stumbled upon using blockchain explorers, forums, or scrolling through a NFT marketplace. Finding something shouldn’t take detective hours of clicking through transaction lists and hash after hash.

The Map arranges blockchain addresses into a 2D view according to their on-chain activities.

Each point in The Map is a blockchain address, and addresses are close to each other if their on-chain activities were similar. This can suggest they transact similar ERC-20 tokens, are both traders rather than hodlers, or are both functional addresses handling transactions for a dApp ( in contrast to persistent wallets representing a user, which will also cluster together).

The layout is decided by a neural network. More interestingly, The Map is how this neural network reasons about addresses within the blockchain protocol, i.e. we are viewing the neural network’s embedding of addresses. The x and y axis of the map have no meaning — moving in a direction, say east of a point, is meaningful relative to that point.

Credit of the original diagram to Aidan Lytle, reposted with permission.

The yellow jellybean is the address embedding. A neural network converts ETH addresses into coordinates within this jellybean. The high dimensional jellybean is then squashed into a 2D plane, creating The Map.

Yes, machine learning has a bias for shape rotators.

Using The Map

Go to

themap.ai

The Map’s legend:

When hovering over an address, address stats like networth estimate and hodler score are displayed. A discussion of hodler score is in the next section. Clicking on an address leads to the etherscan page for that address. For example, clicking on 3 dots in the purple peninsula towards south of the continent lead to addresses that are full of shiba and meme coin transactions. We’ll explain why later.

The Map reflects the Ethereum blockchain state up to July 2022, with no knowledge of on-chain activities past that date. For this and other reasons, address stats such as the networth estimate and hodler score are inaccurate. The Map is for education / entertainment only.

A Map of Ethereum

Ethereum. The protocol where fortunes are made and lost.

Ethereum communities have more OGs, innovators, capital, artists and degens than any other. So a map of Ethereum naturally has all kinds of participants, including NFT collectors, dApp addresses, bots, exchange wallets, Inu millionaires, and ordinary users. Below is the annotated map outlining where these participants live.

High res link

The NFT Coast has lots of addresses with long histories (shown by its cyan coloring) and plenty of ENS registrants (shown by prevalence of diamond symbols). NFT bidders show up here. Whenever a wrapped Ethereum transaction is found in their transaction history, it’s most likely a bid or transaction with a NFT marketplace. Since WETH, and not ETH, is used to place bids in OpenSea, LooksRare, and X2Y2.

In The Map, hodler addresses are pink. They maintain large networths over long periods of time. The Hodler score is calculated from both the magnitude of wealth and the duration over which it is held; by cumulating the address networth over time, the score is an estimate of consistently held value. As the shade of light pink turns to a dark purple, the address hodls… harder.

The Hodler Peninsula is named so for its concentration of hodlers. The majority belongs to one Inu community or another — the leading project of this region being Shiba Inu. So not only was there a significant increase in wealth, but members kept that wealth for longer than expected within the ETH ecosystem.

Distinct from hodlers are high networth addresses at the moment this snapshot of the Ethereum blockchain is taken (July 2022). High networth addresses may have hodled their way to riches, or not. For example, flipping NFTs and frequent trading are other ways to high networth. The Capital is a concentration of wealthy addresses. DeFi tokens such as Curve, Sushiswap, and Aave are overrepresented here among members.

Now we dive deeper into the respective regions.

Region 1: NFT Coast

High res

Besides owning disproportionate number of NFTs and NFT-related erc-20 tokens (LooksRare Token, ApeCoin, x2y2 token), the universal sign of NFT Coast membership is consistent transactions of wrapped ETH, WETH. Earlier in our analysis, the prevalence of WETH transactions gave us the impression this coastline was made up of addresses engaged in multi-chain usage, that use or work with bridges and multi-chain protocols.

But to NFT collectors and frequent patrons of NFT marketplaces, the role of WETH is obvious.

The native currency ETH is not compatible with the ERC-20 standard. Using ETH directly with smart contracts lead to problems. As a result, bidding and NFT transactions in NFT marketplaces are based on WETH. A close reading of how WETH is being used in Etherscan transaction histories reveal interactions with OpenSea, LooksRare, x2y2 wallets.

High res

A typical NFT coast member holds some WETH for convenience, with a disproportionate NFT collection. Other prevalent tokens include ApeCoin and SOS (OpenDao), which make sense.

Participants of the NFT coast have a long history of transactions (800+) and disproportionate registration of ENS domains. They are not exclusive to the highlighted area either, ERC-721 ownership is spread throughout the north coast, the northwest bump (highlighted), and the west coast. Wherever diamonds are present (ENS registrants), good chance NFTs can be found. An open question: what is special about the NFT coast that makes it isolated? NFT ownership is common and interspersed elsewhere, what is so special about NFT owners here?

Region 2: Hodler Peninsula

High res

Diamond handing is second nature to meme coin communities.

The purplish hue of the peninsula means it has a concentration of hodlers, who maintain a large networth for long periods of time. The fact that addresses have hodled historically, but ended with a low networth at the time of our blockchain snapshot (shown by the size of their circles), suggests either a liquidation of assets, or a decline in value for meme coins.

We are finding support for projects like:

Shiba Inu, BabyDoge, Dogelon, Bone ShibaSwap, Doge Killer, ___ Inu

Common assets of a member look like:

High res

Will the hodler status displayed by The Map slowly migrate out of the peninsula, and into other communities, or will Shibarmy maintain their crown as premier hodlers years from now?

The hodler purple seems to concentrate in a vein leading out of the peninsula. This suggests members of meme coin communities are diffused into the rest of the ecosystem.

Region 3: The Capital

High res

The most complex, capital concentrated region of Ethereum is the continent’s slight east of center. That’s given away by the commonality of high networth addresses, ENS domains, and famous addresses.

Common ERC-20 tokens:

NFTs, DeFi (Curve, Yearn, Compound, Sushiswap, Chainlink), stablecoins

It’s not easy to discern trends among Capital members. There is a preference for DeFi projects, relative to other regions. Capital addresses also exhibit a love for NFTs, though less so than the NFT coast, if measured by USD networth divided by # of NFTs.

A portfolio from the Capital can look like:

High res

What’s Next

Developing and analyzing The Map has evolved our thinking of the Ethereum ecosystem. We watched communities form their distinct geography with hardliners, participants that dabble between different Web3 use cases, and the organization of automated addresses into their own clusters (some are marked in the annotated map without analysis). To unlock more opportunities for Web3 as a whole, The Map is looking to:

Be helpful

Visitors will be greeted with up-to-date data, with high fidelity when zoomed into an area. A search function can locate / focus onto relevant addresses. Integration with other analytic products, L2s, and metamask.

Be dynamic

Ethereum is a living protocol. It would be cool to have time differentiated views so the evolution of addresses and communities are visualized.

Treat NFTs as a primary digital asset

NFTs contain meaningful text and image data. Our model should account for NFTs.

Know something we don’t? Want to explore opportunities together? Send feedback and suggestions to lihan.the.map@gmail.com

So, until next time! We are always mapping.

Q & A

How much data was used?

Our neural network was trained using ~6.5 million Ethereum addresses. The Ethereum blockchain snapshot, or the transactions observed, is in the time range of Jan 2020 — July 2022. A lot of addresses weren’t interesting, so we filtered down to 300,000 addresses for display in The Map. In hover text when your cursor is over a point, if num_transactions = 1000, the address most likely have more than 1000 transactions.

I hate this. I want my address delisted!

We believe in the future, being visible on The Map will be an asset to Web3 users and rarely a liability. Analytics for an address can be leveraged by dApps that an user is interacting with, or to support an address’s reputation. For example, doing good accounting of a community’s size, and basing community membership from on-chain activity, is possible.

To delist from The Map, we request proof of address ownership and 2–3 days after request is verified. There is no cost to delist aside from a transaction fee. Please send an email to

lihan.the.map@gmail.com

With the headline “Delist [ 0xYourAddress ]”. In our reply message, a 0 ETH transaction from [ 0xYourAddress ] to our specified address will be requested.

You have a connect wallet button

It dosent do anything, but ‘Connect Wallet’ and a metamask pop-up is too iconic to not include.

--

--

The Map
The Map

Responses (1)