Web-of-Wallets
Go from wallet address level to user level
Unlike IRL and in Web2 where there's a unique ID for each individual, there isn't a universal one in Web3 where it's costless for someone to generate a brand new wallet address. This presents a significant challenge to ensure data integrity and profile data quality.
While anonymity and the convenience of spinning new wallet addresses are "features," they can allow misconduct such as scams, attacks, harmful bots especially in environments without any consequences or regulations.
WoW is one solution to help alleviate these problems by recursively tracing all the historical transactions and behaviors around a given wallet address.
In technical terms, we built a proprietary machine learning algorithm that takes wallet address A as input and produces other wallet addresses that belong to the same owner as output.
- 1.Profile enrichment for going from wallet-level to user-level attributes on gmDATA
- 2.Weed out farmers or fakers in WL token allocation process
- 3.Increase credit risk security
- 4.Get at true # of users for dApps and communities
- 5.Aid rug or shady act investigation
Here's model version 3.11 and its out-of-sample performance.
- Approximately 90,000 pairs of real data
- 8,300 of Unique wallet (A)
- 36,000 of Unique wallet (B)
This model is around 100th iteration. There are other models with 90+ accuracy but we chose this one since it's more balanced and generalizes better out-of-sample.
In practice, Atadia will start including 1 wallet with the highest probability first, then we will scale to 2nd, 3rd, and so on as the model improves even further.
Last modified 2mo ago