What fighting spammers in Beeline Anti-Fraud looks like through a Data Scientist's eyes
summary
Beeline’s Antispam team (part of Anti-Fraud) describes building a service to protect subscribers from unwanted and advertising spam calls. The talk covers data-scientist work: detection models and analytics for identifying spammers, integration with telecom systems to block or label calls, and metrics/operational processes used to improve and monitor the service.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:09What is spam and fraud
- 4:41Spam examples and categories
- 6:06Anti-spam market approaches
- 8:12Beeline traffic volume strengths
- 10:23Number classification difficulties
- 11:47First baseline model
- 14:58Data labeling problems
- 17:25Labeling at number level
- 19:10First baseline with external labeling
- 21:18Optimal activity period
- 23:03Dataset improvement overfitting
- 24:06Final features and models
- 26:31Scoring and blocking process
- 27:58Model monitoring validation
- 30:43How spammers adapt
- 33:12Model improvement plans feedback
- 35:20Summary
- 36:01Q&A
related talks
23mFighting phishing and spam with online machine learning on data streams
59mCalculating by IP: How to Fight Spam on a Social Network
42mDoes Spam have an opinion?
27mIs that spam in my ham? A novice's inquiry into classification.
16mLeveraging ML to Create a Reliable Anti-Fraud System
13mBehavioral Analytics for Real-Time Fraud Prevention
22mDavid Talby, SVP Engineering, Atigeo
41mUsing Machine Learning to Fight Fraud