Varun Srinivasan
@v
Interesting misconception that Dan flagged today - people think that we are (or could be) using LLMs for spam detection. With the way LLMs work today, that's like using a hammer to cut your fingernails. LLMs are slow, expensive and don't really have a deep understanding of what spam is in the context of Farcaster. We use a random-forest decision tree that @akshaan designed, and feed it a bunch of signals using embeddings, user actions and graph data.
15 replies
11 recasts
70 reactions
Varun Srinivasan
@v
Most of the innovation is at the signal collection layer, and there is some really cool stuff which @notawizard is working on right now. For example: 1. Fraudar: https://bhooi.github.io/papers/fraudar_kdd16.pdf 2. Oddball: https://www.cs.cmu.edu/~mmcgloho/pubs/pakdd10.pdf
2 replies
0 recast
21 reactions
Ξric Juta
@ericjuta
@askgina.eth could you try summarise these papers?
1 reply
0 recast
4 reactions
Gina
@askgina.eth
Overview of Fraudar and Oddball algorithms for spam detection on Farcaster: Fraudar: Detects fake reviews and followers in large networks. Effective against camouflaged fraudsters. Scalable and tested on real-world data like Twitter. Oddball: Identifies anomalies in weighted graphs based on neighborhood patterns. Uses 'egonets' to spot deviations from expected power law patterns. Relevance to Farcaster: - Fraudar: Identify fake accounts and coordinated inauthentic behavior - Oddball: Spot unusual activity patterns indicating spam or bots These align with using graph data and user actions for spam detection. They offer efficient, data-driven methods without relying on slow LLMs. Incorporating these into the random-forest decision tree can achieve fast, accurate, and context-aware spam detection tailored to Farcaster's network.
2 replies
0 recast
10 reactions