Building reputation vectors using honeypot profiles on Facebook

Building reputation vectors
using honeypot profiles on
Facebook
Nasri Messarra, Anne Mione
Université de Montpellier 1 - MRM
1

Background
 E-reputation has become an important concern for firms
 Pampers, Nestlé and other brands have already paid the heavy price of fan
attacks (Champoux et al., 2012; Paul Gillin, 2012; Steel, 2010).
 The observation of the buzz and more particularly of the negative buzz (bad
buzz) is important (Cuvelier, Aufaure, 2011)
 Attacks on Facebook are more frequent and research is required to better
understand and counteract them
2

Litterature review 1 : E-reputation
 1990 : Howard RHEINGOLD creates the concept of « e-reputation ». He
evoques a « digital social life »: public debates, controversies have an impact
on the reputation of the firm, individuals, associations, etc…
 2000 : appearance of tools to manage e-reputation
 customer ratings (evaluation of the supplier by customers on Ebay auctions,
evaluation of books by customers on Amazon, etc.)
 and positive and negative buzz management :
 contribute to positive buzz and
 identify and counteract negative buzz.
3

Litterature review 1 : E-reputation
 E reputation)
Three theoretical ambitions (François Xavier de Vaujany, Helène Lambrix, DRM) :
 holistic ( the organisation is the level of analysis)
 Individualist ( the models are centered on individuals or a key individual in or
outside organization)
 Integrative ( holistic and individualists aspects are simultaneously taken into
account)
Litterature concerns the shaping (distortion) of reputation: antecedents,
maintenance and reputation effects (Deephouseet Carter, 2005 ; Rindova et al.,
2005 ; MacMillan, 2007 ; Walsh et al., 2007 ; Lange et al., 2011)
 We consider a destabilizing action on reputation
4

Litterature Review 2 : Viral Marketing5
 Hinz, Skiera, Barrot & Becker (2011) define 4 critical viral marketing success factors:
 1. Content, in that the attractiveness of a message makes it memorable (Berger and
Milman, 2011; Berger and Schwartz, 2011; Gladwell, 2002; Porter and Golan, 2006)
 2. The structure of the social network (Bampo et al. 2008)
 3. The behavioral characteristics of the recipients and their incentives for sharing the
message (Ardnt, 1967);
 4. The seeding strategy which determines the initial set of targeted consumers chosen
by the initator of the viral campaign (Bampo et al., 2008; Kalish, Mahajan, and Muler,
1995; Libai, Muller and Peres, 2005)

6 Litterature Review 3 : Seeding strategy
There are debates in literature concerning seeding population :
- Target hubs is not sufficient (Watts et Dodds, 2007, Skiera, Barrot, Becker, 2007)
- Target hubs is still important (Hinz et al. , 2011)
- Size does not matter (Scarpi, 2010) and one of the keys to success is the initial seeding population
(Liu-Thompkins, 2012)
-Literature refers to fans and community members as “nobodies” and “somebodies” (Booth & Matic,
2011) and more and more researchers focus on the quality of the members rather than on the size of
network (Scarpi, 2010; Wallace, Buil, Chernatony, 2014).

Research questions
 Can we conceive and realize an original attack of a company through Facebook
 in real situation but with the General Management agreement (Research action)
?
 In order to :
 Check the feasability of such a strategy
 Increase the methodological understanding of such an attack
 Measure the WoM diffusion model
 Contribute to the seeding population strategy debate
7

Methodology
 we target a small European company with around 1,500 fans on its Facebook page
 European market
 Services activity
 We create a schema
8

Methodology: How consumers act on Facebook with or
against brands and companies
1- Post directly on the
brand’s page
2- Post on their own social networks and wait
for the viral effect of WoM to reach the brand
Reported cases & littérature about Nestlé, Pampers, DKNY,
Marie-Claire, Capri Sun, Cooks Source, Bershka…
9

Personal
timeline
Brand community
On Facebook
(Brand page)
A three steps attack through OSN using an
optimized seeding population
1. Create a fake profile
2. Attract engaged fans of a brand to
befriend the fake profile (our initial
seeding population)
3. Diffuse information organically to
engaged fans (no need for WOM for
diffusion)
Engaged
fans
10
Methodology: The schema

First step : Create a Fake profile
 Based on homophily (cover experiment)
 Stereotype of existing fake profiles on Facebook (Barracuda networks, )
11

Cover Experiment
29
39 44
63
78
113
50
75
101
131
164
180
195
0
50
100
150
200
250
0 22 24 33 37 70 209
Friends
Day
Cover Experiment
Requests Sent / Requests Accepted
Acceptance Request sent cumulative
12 Second step : acquiring friends

Stereotyped fake Profile
Barracuda Networks’ statistics (Rashid, 2012): a
woman living in a major city, having a high
education degree, interested in both men and
women…
We attract 200 engaged fans
in 7 months
13 Second step : acquiring friends

14
Second step : acquiring friends
1. We visit the brand’s Facebook page
2. We locate the fans who engaged with the brand (like, comment, share)
3. We send a friend request from both profiles a week apart

 We reuse 6 posts from the fan page of the brand
15
Third step : diffusion of information

~200 engaged
fans as friends
from stereotype
~100 engaged
fans as friends
from cover
Only 30 mutual friends
16
Results : resulting network

 In average, 105 engaged fans receive the message organically
 44 “friends” engage in the conversation
 Generates WoM and interaction
17
Results: diffusion of information and generation of WoM

Results: Directed Communication
Graph of All Conversations
 Both profiles generated one larger network of communication
and influence.
 They acted as bridges between two sub-networks who react
differently (we mentioned that our profiles only have 30 friends
in common
 33 engagement actions (comments/like) where made on the
statuses or comments posted by one of the honeypot profiles
and 44 engagement actions were made on the statuses or
comments made by the other honeypot profile.
 14.8% of edges (connections) are reciprocated showing that
communication got back and forth between engaged fans.
 The maximum geodesic distance is 5 showing that the
information was viral to a certain extent and travelled from one
node to the other with an average of 2.32 nodes on a path.
18

Results
 Before and After
19
Negative posts published
on the brand's Facebook
page
Same negative posts
published on fake
profiles timelines using
an optimized seeding
strategy
Difference
Fans engaged 11 37 +236%
Engagement actions (likes,
comments)
15 77 +413%
Reciprocation (back and
forth communication and
consumer to consumer
communication)
0% 14.8% +14.8%

Results
 Posts were directly received by engaged fans of the brand itself which shared,
commented or liked these posts, generating word of mouth reaching friends of
friends to the 5th level
 Information was served directly to these fans at no cost (organically) and at a
higher reach than it would have been possible if these posts were published
directly on the company’s page.
 This forced changes IRL (in real life)
20

Results
 This answers our initial question and confirms our hypothesis: it is possible to attract
engaged fans of a brand and distribute information organically to them bypassing the
control of the brand itself and preventing it to stop the information sharing.
 The campaign was a success: The company did not take any action for a while and
stood as an observer until it realized that the movement was not going to stop and
that it would not be able to contain it. It then tried to respond shyly to our posts and,
finally, changes were made to the company’s management team.
21

Back to the 4 critical viral marketing success factors (Hinz, Skiera, Barrot & Becker (2011):
1. Content (Berger and Milman, 2011; Berger and Schwartz, 2011; Gladwell, 2002; Porter and Golan, 2006)-
We did not work on the content, using posts already published on the brand’s page
2. The structure of the social network (Bampo et al. 2008): We engineered a network of engaged fans of a brand
around fake profiles and described a schema that can be reproduced or anticipated.
3. The behavioral characteristics of the recipients and their incentives for sharing the message (Ardnt, 1967):
We helped reach more engaged fans without dispersion using organic reach (without WoM) which resulted in a higher
engagement and interaction.
4. The seeding strategy (Bampo et al., 2008; Kalish, Mahajan, and Muler, 1995; Libai, Muller and Peres, 2005): We
showed that an optimized initial population engages more with posts , which results in higher pressure and
influence offline.
22
Contributions to Viral marketing, WoM Literature

Contributions to Social Network building
methodology
 We engineered a network of online “friends” who may not be friends or
even know each other in real life. This social network would have never
existed if it wasn’t for our experiment.
 We created a new type of Honeypot profiles
 Webb, Caverlee and Pu (2008) defined Honeypot profiles as real profiles used to attract fake ones.
Our experiment does the opposite as it uses fake profiles to catch real ones.
 attraction realized by homophily:
 Boshmaf, Muslukhov and Beznosov (2011) used social bots to attract friends based on mutual
frienship. Our cover experiment shows that, on Facebook, people may engage others based on
homophily:
23

Contributions to e-reputation
management
 Managers should be aware about the usage of such vectors that could
ruin the reputation of a firm.
24

Discussion
 We are aware that our experiments on honeypot profiles and influence in online
social networks raise ethical questions.
 The number of fake profile is so important nowadays that scholars have to develop
their knowledge about them, as they constitute a potential tool in influence
strategies.
 The same experiments can be reproduced with real Facebook accounts and the
findings can be used as well to create engaged communities and improve a
company’s reputation or promote brands in an ethical way.
25

Conclusion
 Many scholars agree that the key to success in information diffusion is influencing
the influencer (Galeotti & Goyal, 2009; Hinz et al., 2011) and using strategy to
target an optimized initial seeding population (Liu-Thompkins, 2012).
 Information of peers tends to be more influential than the information diffused by
brands or similar source (Hinz et al., 2011).
 Our contribution exposed a strategy to build an optimal seeding population from
scratch in a container, the circle of friends of our fake profiles, and diffuse
information organically to them.
26

Thank You
 Nasri Messarra
 Anne Mione
27

Building reputation vectors using honeypot profiles on Facebook

More Related Content

What's hot (19)

Viewers also liked (11)

Similar to Building reputation vectors using honeypot profiles on Facebook (20)

More from Nasri Messarra (7)

Recently uploaded (20)

Building reputation vectors using honeypot profiles on Facebook

Editor's Notes