Bot or Not

Using data analysis we first train a machine learning model to uncover the some insight into the bot activity on Lens Protocol by generating a publicly available dataset of over 100k profiles. However as is always the case, nothing is perfect. Every model is biased and it is impossible to build mindblowing models in few hours. Furthermore labeling the data is the most time consuming and often iterative process. Here in this project we deploy the model for both users to explore the ecosystem but also help to better differentiate between the human and bot activities. We store the user contributions on-chain (Scroll Alpha testnet) and paying them the transaction. We think that this provides users incentives to contribute to the ecosystem as it leaves permanent traces. Furthermore the user inspection is made easier by providing additional Token/NFT data from Ethereum Mainnet that was not used in the training of machine learning. This is done with the incredible support Airstack.

This solo hacked project combined data analytics (scraping like hell for Lens Protocol + onchain data). ML/AI using Google tabnet semisupervised learning (oldie but goldie) for bot classification. Gaming UI to incentivize people to contribute to labeling data and potentially validating the results. New GraphQL checking with Airstack. Cool resource and especially GPT4 plugin to write faster queries. Although complex queries still need some work. Fast data offers lot of interesting possibilities! Here we use it as a tool to assist analytics. Deploying attestations of user contributions to Optimism AttestationStation contract that this project deployed to Scroll. Besides these. By far the biggest resource that was hacked was Lens Protocol. Flipside Crypto was used for additional analytics. BigQuery, NextJS, Python, NodeJs, Foundry for contract deployment/verification, Jupyter Lab, ChatGPT for collaborative coding and of course Pocketwifi.