Fighting against Marketplace Scams using Community-driven AI
summary
Claims a production pipeline that ingests Discord community reports and message metadata, stores engineered features in a feature store, and scores messages in real time with pretrained Transformer models (BERT/GPT) plus simple heuristics (account age, follower count, mention counts, scam phrases, negative reactions). Uses community-reported labels to continuously relearn; OpenSea-style fraud prevalence cited (~80% spam), and the system both flags scams in real time and surfaces educative warnings to reduce future abuse.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:04Feature engineering overview
- 3:12Why focus on data
- 5:39Marketplace scams explained
- 7:51Digital vs physical scams
- 8:55Discord communities as data source
- 12:03Case study Twitter scam detection
- 14:07Defining scam features
- 16:17Feature engineering for messages
- 18:45Semantic features with language models
- 19:07Role of feature stores
- 20:34Community-driven AI solution
- 21:38Call to action and conclusion
related talks
25mUnsupervised shallow learning for fraud detection on marketplaces
16mLeveraging ML to Create a Reliable Anti-Fraud System
13mHarnessing Machine Learning to Combat E-commerce Fraud
38mEthical Artificial Intelligence to Advance Welcoming Communities and Identify Potential Communica...
42mBuilding a Robust Fraud Detection System
24mEngaging new and diverse audiences with ML
16mFighting Harmful Content with AI: Real-World Safety
35mGenerative AI for Fraud Detection & Claims Analysis