mining social networks using heat diffusion processes for marketing candidates selection
DESCRIPTION
Mining Social Networks Using Heat Diffusion Processes for marketing candidates selection. Hao Ma, Haixuan Yang, Michael R. Lyu and Irwin King CIKM, 2008. Reported by Wen-Chung Liao, 2009/10/6. Outlines. Motivation Objectives Diffusion models - PowerPoint PPT PresentationTRANSCRIPT
1Intelligent Database Systems Lab
國立雲林科技大學National Yunlin University of Science and Technology
Mining Social Networks Using Heat Diffusion Processes for marketing candidates selection
Hao Ma, Haixuan Yang, Michael R. Lyu and Irwin King
CIKM, 2008.
Reported by Wen-Chung Liao, 2009/10/6
2
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Outlines
Motivation Objectives Diffusion models Marketing candidates selection algorithms
and their complexity Empirical analysis Conclusions Comments
3
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Motivation
Due to the complexity of social networks, few models exist to interpret social network marketing realistically.
Studies of innovation diffusions, they were descriptive, rather than predictive─ they are built at a very coarse level, typically with only
a few global parameters─ and are not useful for making actual predictions of the
future behavior of the network.
4
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Objectives
Model social network marketing using Heat Diffusion Processes.
Presents three diffusion models, along with three algorithms for selecting the best individuals to receive marketing samples.
5
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.HEAT DIFFUSION MODELS The process of peopl
e influencing others is very similar to the heat diffusion phenomenon.
In a social network, the innovators and early adopters of a product or innovation act as heat sources.
6
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Diffusion on Undirected Social Networks G = (V,E)
G = (V,E)• V is the vertex set, and V = {v1,
v2, . . . , vn}.• E is the set of all edges (vi, vj ).• fi(t) describes the heat at node vi
at time t• fi(0) initial value. • f(t) denotes the vector consisting
of fi(t).• M(i, j, Δt): amount of heat from n
ode j to node i during a period Δt
7
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
Diffusion on Directed Social Networks
Diffusion on Directed Social Networks withPrior Knowledge
8
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
O(N(PM+N +N logN))
O(N(PM +N +kd))
O(kN(PM +N +d))
Marketing candidates selection
9
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.EMPIRICAL ANALYSISEpinions• maintains a “trust” list which presents a network of tr
ust relationships between users,• product categories, “Kids & Family”• 75,888 users, and 508,960 edges• the initial heat vector f(0), choose N/k• the thermal conductivity value α, set α= 1• the adoption threshold θ, set θ = 0.6• t = 0.10, t = 0.15 and t = 0.20, unit???Scenario: • 1 to 20 product samples (k =20) • the marketing candidates? • performance (measured by the value of coverage) ?
10
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.EMPIRICAL ANALYSIS
11
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Conclusion
Propose a social network marketing framework which includes three diffusion models and three marketing candidates selection algorithms.
Model social network marketing as realistically as possible
Can defend against diffusion of negative information,
This framework is scalable.
12
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Comments
Advantage─ Realistic & scalable.─ Defend against diffusion of negative
information Shortage
─ Static social network. My opinion: