1. Competition in VM – Completing the Circle

2. Previous work in Competitive VM Mainly follower’s perspective: given state (say of seed selection) of previous companies (agents/players): – what’s the best strategy for the “follower” to maximize its spread in the face of the competition? – What’s the best strategy for the follower to maximize the blocked influence of the opponent? Most competitive VM algorithms not scalable or assume unfettered access to the n/w for all players. – How realistic is that?

3. But … Campaign runners don’t necessarily have unfettered access to the network! There is an owner of the network. Campaigns need owner’s permission. May need to pay the owner.

4. A New Business Model – Introducing … Network owner Provides VM service. How should the host select/allocate seeds? I need 100 seeds I need 250 seeds Competition starts after host selects/allocates seeds.

5. Business Model (contd.)

6. Why is fairness important? Imaginary scenario: SONY NEX-VG30H 50 seeds Spread 1000 JVC HD Everio GZ-VX815 30 seeds Spread 240 For comparable products, if the b4b is substantially different, dissatisfied company(ies) may take their VM business elsewhere!

7. Propagation Model Which model should we use? (C)LT – more natural for product adoption. Unfortunately, previous CLT proposals have some disadvantages. – The CLT discussed last class (abstracted from Chen et al. and Borodin et al.) – not submodular. – Spells computational difficulties. Let’s review Borodin et al.’s model next.

8. WPCLT Model inactive influenc ed active Should I buy a camcorder? Which CC should I buy? I’ve chosen my color! Seeds influence out-neighbors (followers). Active nodes influence followers. Once influenced, relative weights of different (campaign) influences induce a random decision making trial.

9. WPCLT Model Core – LT Model.

10. WPCLT Analysis u 0.5 v 0.5 w 1.0 x

11. WPCLT Analysis u 0.5 v 0.5 w 1.0 x Red seed happens to help blue’s cause!

12. Introducing the K-LT Model

13. Proof of both properties

14. Fine, what do we want to optimize?

15. Optimal Seed Selection

16. Fair Seed Allocation

17. Calculating Spread/Marginal gain Adjusted Marginal Gain 0.4 0.3 0.50.1 0.2 u v w

18. How hard is FSA?

19. Illustrating the Reduction

20. FSA remains NP-complete but weakly so. It resembles PARTITION in this case. Can be solved in pseudo-poly time using dynamic programming. Efficient heuristic for general case: Needy Greedy!

21. Needy Greedy Algorithm

22. Experimental Evaluation & Conclusions EE: See paper. Role of host. Fair seed allocation. Complexity (strongly NP-complete) and efficient heuristic solution. Questions: What if the influence weights and node thresholds depend on the company? Other sort orders for seeds in Needy Greedy? What if the products significantly differ in popularity: e.g., fairness achievable b/w Nokia and Galaxy is limited. How do we handle this? Adaptive seed selection/allocation (game theory)? Sometimes propagation phenomenon may proceed like an “S-curve” – need “critical mass” of seeds to reach tipoff point. Blind pursuit of FSA here may be problematic.