Gong info heist
0 likes448 views
The document discusses planning a heist by analyzing information flow between individuals. It outlines four phases of a heist: the mark, the team, the plan, and the takedown. For analyzing information flow, it discusses challenges like hidden networks and subtle signals. It proposes using graphical causal models, mutual information theory, and large datasets to study when information flow can be inferred, the direction of flow, and the quantity of flow. The goal is to develop a framework for testable theories of information spread.
![The mark: Information 鍖ow
Under what conditions can we infer...
1. that information has 鍖owed among people?
2. the direction of information 鍖ow?
3. the quantity of information 鍖ow?
To speak with precision about [information 鍖ow] is a task
not unlike coming to grips with the Holy Ghost.
- V. O. Key, Public Opinion and American Democracy](https://image.slidesharecdn.com/gonginfoheist-111128100502-phpapp01/85/Gong-info-heist-9-320.jpg)
Gong info heist
- 1. How to plan a heist: Challenges, models, and tactics for making inferences about social information 鍖ow Abe Gong CSAAW - Nov. 2011
- 2. Heists!
- 3. Phases of a heist 1. The mark A powerful, dangerous enemy who deserves to be taken down
- 4. Phases of a heist 1. The mark A powerful, dangerous enemy who deserves to be taken down 2. The team A group of mis鍖ts and outcasts with diverse talents
- 5. Phases of a heist 1. The mark A powerful, dangerous enemy who deserves to be taken down 2. The team A group of mis鍖ts and outcasts with diverse talents 3. The plan Manipulates assumptions and information to get through the marks defenses
- 6. Phases of a heist 1. The mark A powerful, dangerous enemy who deserves to be taken down 2. The team A group of mis鍖ts and outcasts with diverse talents 3. The plan Manipulates assumptions and information to get through the marks defenses 4. The takedown The plan is executed and all surprises are revealed
- 7. The mark
- 8. The mark: Information 鍖ow Under what conditions can we infer... 1. that information has 鍖owed among people? 2. the direction of information 鍖ow? 3. the quantity of information 鍖ow?
- 9. The mark: Information 鍖ow Under what conditions can we infer... 1. that information has 鍖owed among people? 2. the direction of information 鍖ow? 3. the quantity of information 鍖ow? To speak with precision about [information 鍖ow] is a task not unlike coming to grips with the Holy Ghost. - V. O. Key, Public Opinion and American Democracy
- 10. The mark: Challenges Hidden networks: We dont know where people get their information. Subtle signals: Even when we know where the information comes from, we dont know how people process it. Our theories are grossly underspeci鍖ed.
- 11. The team
- 12. The team Judea Pearl Graphical models of causality
- 13. The team Judea Pearl Graphical models of causality Claude Shannon Information theory, esp. measurement
- 14. The team Judea Pearl Graphical models of causality Claude Shannon Information theory, esp. measurement Mark Zuckerberg Lots and lots of data
- 15. The team: Pearls graphical causal models Correlation implies some kind of causation. AB AB or B A or C {A, B} Graphical models let us pin down knowns and unknowns. d-separation allows us to ignore the rest of the network.
- 16. The team: Shannons mutual information Crisp, general measure of shared information. p(x,y ) I (X ; Y ) = y x p(x, y )log ( p(x)p(y ) ) Works on conditional probabilities as well. Works on individuals and ensembles allows aggregation. Provides a nice framework for discussing social in鍖uence.
- 17. The team: Zuckerbergs mountains of data Lots of data about lots of people Includes text and other high-bandwidth signals Includes time stamps, and directed links
- 18. The plan
- 19. The plan: Objectives Goal: A framework (axioms and notation) for testable theories of information 鍖ow. When can we infer... 1. that information has 鍖owed among people? 2. the direction of information 鍖ow? 3. the quantity of information 鍖ow?
- 20. The plan: Existence of 鍖ows Pearl (solo): Correlation implies (some kind of) causation. Examples 1. Plagiarism 2. Newton and Leibnitz 3. Surges in google trends
- 21. The plan: Direction of 鍖ows Pearl: Experiments, when possible. Zuckerberg: Action space mining Pearl and Zuckerberg: Timestamps and poor mans causality Examples 1. Canary trap 2. memetracker 3. retweets 4. Christmas tree sales
- 22. The plan: Size of 鍖ows Shannon and Zuckerberg: behavioral aggregation Group similar actors and assume they respond to information in the same way. Allows us to parameterize f (). Shannon and Pearl: causal aggregation Group similar actors and assume they are receiving the same information Makes more parts of the network measurable. Shannon, Pearl and Riolo: simulation Group similar actors so that all important info sources are measureable. Examples: 1. Convention bumps in political campaigns 2. ...?