an agent-based simulation of viral marketing effects in...

An Agent-Based Simulation

of Viral Marketing Effects in

Social Networks

Axel Hummel1, Heiko Kern1, Stefan Kühne1 and Arndt Döhler2

1 Business Information Systems, University of Leipzig 2 Intershop Communications AG

ESM'2012 –

26th European Simulation

and Modelling Conference

Outline

Introduction

Simulation model

The Facebook domain

Model structure

Behaviour of the agents

Simulation results

Conclusion and future work

Axel Hummel 2 22.10.2012

Motivation


Source: http://www.squaremartinimedia.com/wp-

content/uploads/2010/10/facebook-friends-box.jpg

Social networks

connect friends

Interesting news are

send to the friends

Viral effects provide

new marketing

possibilities Source: http://lh4.ggpht.com/-

W1j4mm_HymI/Tem3N0f_B1I/AAAAAAAAB2o/l6TIAiPd

mvU/Viral-Marketing%2525255B4%2525255D.png

Problem


Companies have a lack of

experience with social networks

They do not know how to use social

networks for viral marketing

Return on investment calculation

is a challenging task [1]

? ? ?

Solution

We use agent-based simulation to forecast

the effects of marketing campaigns

The goal of the simulation is to answer the

following questions

1. How many Facebook fans are expected?

2. What are the benefits for the online shop?

3. What costs are incurred?


The Facebook domain


Facebook

User

Company

Facebook

User

Facebook

User

Facebook

User

Facebook

User

isFriend ►

isFriend ►

The Facebook domain


Facebook

User

Company

Facebook

User

Facebook

User

Facebook

User

Facebook

User

isFriend ►

The Facebook domain

Two main requirements for the simulation

model

(1) Representation of the individual

friendship relationships

(2) Modelling of the communication process

between

A company and its fans

The Facebook friends


Overview of the model structure


onlineActivity : OnlineActivityuserType : UserTypeisRegularCustomer : Boolean

regularCustomerBecomeFan : DoublememberBecomeFan : DoubleperceptionProbabilityNewsfeed : DoubleclickThroughRate : Double

<<agent>>User

name : StringadvertisementActivity : IntegersetupCosts : DoublecostsPerFanPageUpdate : Double

<<agent>>Company

type : NewsTypetime : Integer

News

hasFriends

*

isFriend

*

NewsFeed

*

1

1

0..1

* hasFans

isFanOf

1

* sentNews

isSentBy

1

sentNews

isSentBy*

User agent




<<agent>>User


<<agent>>Company


News

hasFriends

*

isFriend

*

NewsFeed

*

1

1

0..1

* hasFans

isFanOf

1

* sentNews

isSentBy

1

sentNews

isSentBy*

Behaviour of the user agent


Log in

Read news feed

Press

fan button?

Want to visit

Facebook?no

yes Is regular

customer?no

yes

Visit

online shop?

Press

like button?

Log out

Press

fan button?

[!pONCE, if onAc = ONCE][!pSEVERAL, if onAc = SEVERAL]

[false, if onAc = DAILY]

[true, if onAc = DAILY][pSEVERAL, if onAc = SEVERAL][pONCE, if onAc = ONCE] [isRegularCustomer = true]

[isRegularCustomer = false]

Press

fan button

yes

no

[regularCustomerBecomeFan]

[!regularCustomerBecomeFan]

[isFan = false] and

[isFan = true] or

Press

like button

[userType = ACTIVE_USER]

[userType = LURKER]

Press fan

button

yes

[memberBecomeFan]

[!memberBecomeFan]

[isFan = false] and

[isFan = true] or

Visit

online shop

yes

no

[clickThroughRate]

[!clickThroughRate]

[visitShopThisDay = false] and

[visitShopThisDay = true] or

[feed is empty] [feed is not empty]

[NewsType = FRIEND_NEWS]

[NewsType = COMPANY_NEWS]

Every message is read with a probability of perceptionProbabilityNewsfeed,otherwise a message is ignored.

yes

no

no



Log in

Read news feed

Press

fan button?

Want to visit

Facebook?no

yes Is regular

customer?no

yes

Visit

online shop?

Press

like button?

Log out

Press

fan button?





Press

fan button

yes

no



[isFan = false] and

[isFan = true] or

Press

like button


[userType = LURKER]

Press fan

button

yes

[memberBecomeFan]

[!memberBecomeFan]

[isFan = false] and

[isFan = true] or

Visit

online shop

yes

no

[clickThroughRate]

[!clickThroughRate]







yes

no

no

Company agent




<<agent>>User


<<agent>>Company


News

hasFriends

*

isFriend

*

NewsFeed

*

1

1

0..1

* hasFans

isFanOf

1

* sentNews

isSentBy

1

sentNews

isSentBy*

Behaviour of the company agent


Update fan page

Want to update fan page?

no yes

[day % ad != 0] [day % ad = 0]

Model configuration

3 groups of input parameters

Company-specific parameters

Advertisement activity, setup costs, …

Facebook-specific parameters

Daily online rate, lurker rate, …

Social structure of the user agents

Facebook sub network of the University of

Pennsylvania [2]

41,554 users, 1,362,229 friendship relations


Simulation results – growth of

Facebook fans (365 days, 10 runs)


0

1000

2000

3000

4000

5000

6000

7000

8000

9000

0 25 50 75 100 125 150 175 200 225 250 275 300 325 350

Nu

mb

er

of

fan

s

Days

Growth of Facebook fans

Overall fans (average case) Regular customer fans (average case)

Overall fans (worst case) Overall fans (best case)

Simulation results – different

advertisement strategies


0

5000

10000

15000

20000

25000

30000

35000

never every 14 days every 7 days every 3 days every day

Number of fans 4727 4956 5070 5390 6050

Number of shop visits 635 1099 1490 2631 6397

Overall costs (in €) 3000 4950 6900 12075 30375

Number of fans Number of shop visits Overall costs (in €)

Message time interval

Costs per Facebook fan &

costs per online shop visit


0 €

1 €

2 €

3 €

4 €

5 €

6 €

never every 14 days every 7 days every 3 days every day

Costs per Facebook fan and costs per online shop visit

Costs per Facebook fan (€) Costs per online shop visit (€)

Message time interval

Conclusion

Number of Facebook fans and number of

shop visits depend on the message time

interval

A continuous stream of visitors requires

high Facebook activities

This results in high costs


Summary

Agent-based simulation technique is

applied to the social networks domain

Model is calibrated and validated by real

data and the outcome of several studies

Online shop managers can optimize their

Facebook marketing activities


Future work

Limitations

Special marketing events are not considered

(advertisements, prize competitions, …)

The decrease of fans is excluded

Other topics

Refinement of user relationships and their

influence on the user behaviour (trust model)

Refinement of the EdgeRank Algorithm


References

[1] Intershop Communications AG, 2011. SimProgno study:

Decision Support in E-Commerce.

URL: http://simprogno.de/downloads/Auswertung_Simprogno_Studie_2011.pdf.

(in German).

[2] Traud A.L.; Mucha P.J.; and Porter M.A., 2011. Social

Structure of Facebook Networks. CoRR, abs/1102.2166.


http://simprogno.de/downloads/Auswertung_Simprogno_Studie_2011.pdf

Thank you for your attention!

Contact information:

Axel Hummel

Business Information Systems

University of Leipzig

Augustusplatz 10

04109 Leipzig, Germany

phone: +49 341 9732303

[email protected]

http://bis.informatik.uni-leipzig.de/AxelHummel

The SimProgno Project

http://www.simprogno.de

Funded by the German Federal

Ministry of Education and Research


mailto:[email protected]






http://www.simprogno.de/

an agent-based simulation of viral marketing effects in...

Documents