[Meeting Notes CX] Weekly SW3

[nod0n]

What are the most important topics discussed?

• what fraud do we analyze?
  • exit scams / premine scams
  • pump and dump
• where do we store scraped data
• what kind of data and where do we scrape?

Most important insights since last weekly

• There are different kind of fraud detection in the crypto space:
  • Which coin is a fraud?
  • Which single transaction is fraud?
  • Fraud with smart contracts
• Pump and dump happen on small coins that are easier to manipulate and on single exchanges and take only a couple of minutes.

What was decided?

• Topic's -> what kind of fraud to analyze?

  • exit scams / premine scams

• Scraped data -> store in Elastic Search

Next 2-3 weeks

• Define coins (Base)

  • time frame of interest
  • Coins:

    • Scam:

      • FTX Token
      • Terra Luna Classic
      • Squid game
      • STA
      • BeerCoin
      • BitForex

    • No Scam:

      • Bitcoin
      • Ethereum
      • Cosmos
      • AVAX

• Proof of concept for scraping (for WDB)

  • How much data can be gathered per day/hour
  • which data sources (selenium)?
    • Reddit as website (FA)
    • X as website (CG)
    • Telegram as website (FA)
    • Google Results (CG)
    • Discord? (FA)
    • Comparitech Rug Pulls List (CG)
    • Google Trends (FA)

• Scrape data

besprechung moritz

• paper:
  • klar definieren auf welches fraudthema wir uns konzentrieren
  • dass Modelle keine Blackbox werden:
    • expl-ai
    • zeitreihen: einstieg preis hoch/tief -> modelle validieren
    • modelle müssen statistisch verheben
• oversampling -> für ausgewogenheit fraud / nicht fraud
  • learningrates ändern (Möglichkeit)
  • variablität von no scams zeigen
• pro scamCoin automatisch ein gleichwertigen normalCoin suchen -> dass es statistisch verhebt
  • liste aufstellen, welche etwas gleiche Kapitalisierung haben

ziel idee

• mehrere Modelle
  • von social Media
  • von Preisdaten
• modelle zusammen zeigen fraud prop.

feelings

all members are motivated :)