The Use of a Fuzzy Cognitive Maps and Eye Tracking

in Exploitation of Online Advertising Resources

Jarosław Jankowski

, Jarosław Wątróbski

, Katarzyna Witkowska

and Waldemar Wolski

West Pomeranian University of Technology, Żołnierska 49, 71-210, Szczecin, Poland

University of Szczecin, Mickiewicza 64, 71-101, Szczecin, Poland

Keywords: Fuzzy Cognitive Mapping, Online Marketing, Eye Tracking.

Abstract: Current trends in the area of digital marketing and online promotion indicate the increasing role of highly

targeted online advertising. The method of designing advertisements evolved in the direction of one-to-one

communication, which forces to create personalized advertising messages. In the case of interactive media

modelling changes can occur in a dynamic way based on audience and content characteristics. The purpose

of this article is to develop a model supporting the selection of parameters and localization of advertising

content with the use of fuzzy cognitive maps and possibility to perform simulations based on content

selection towards increased efficiency.

1 INTRODUCTION

Nowadays in marketing different promotional

instruments and measures, with which a company

communicates with the surrounding, are used.

The increase in the importance of interactive

technologies, especially of the Internet

in an organization’s marketing activity, is related,

among other things, to new media, which emerged

thanks to the development of information

technology (Kotler and Postma, 1999). Functions of

the traditional media are being taken over by the

electronic media or their combination in the form of

multimedia. The interactive media with one-to-one

model allows to send an announcement to a recipient

and receive feedback (Hoffman and Novak, 1995)

what is a result of evolution of one-to-many

communication model from traditional media (Rust,

1989). The basic feature, which distinguishes the

electronic media (including the Internet),

is a possibility of employing bidirectional

communication which led to the development

of new communication models (Hoffman and

Novak, 1995). The way of designing advertisements

is moving in the direction of targeting (Goldfarb and

Tucker, 2011), maximizing the level of

personalization with the use of morphing techniques

(Theocharous and Thomas, 2015), usage of ad

recommendation systems (Theocharous and

Thomas, 2015) or the use of characteristics derived

from social network connections (Wan-Shiou et al.,

2006). In the case of modelling an interactive

message, its changes can be dynamic and on the

basis of observations one can systematically

introduce changes in order to compare different

concepts and designs (Theocharous and Thomas,

2015). Assisting decision-making processes

concerning broadcasting a spot is of great

importance to Internet advertising campaigns.

A number of factors make it difficult to choose

a proper criterion for activity optimization.

Presented in this paper approach is based on the eye

tracking and advertising content representation

based on fuzzy cognitive maps (FCM). In modelling

with the use of FCM one needs to take into

consideration the fact that prepared characteristics

are in most cases based on human knowledge.

In order to avoid subjective opinions the paper

proposes the solution which applies eye tracking

to calculate relations between the elements

of and parameters of examined advertising content.

Paper is organised as follows. Section 2 introduces

theoretical background concerning tools used

to construct a model, that is fuzzy cognitive maps

as well as the eye tracking methodology. Section

3 shows authors’ example of employing cognitive

maps in modelling an interactive message

and presents the description of the experiment with

the use of an eye tracker. Section 4 discusses

the achieved results and is followed by Summary.

Jankowski, J., W ˛atróbski, J., Witkowska, K. and Wolski, W.

The Use of a Fuzzy Cognitive Maps and Eye Tracking in Exploitation of Online Advertising Resources.

In Proceedings of the 18th International Conference on Enterprise Information Systems (ICEIS 2016) - Volume 2, pages 467-472

ISBN: 978-989-758-187-8

467

2 MATHEMATICAL

REPRESENTATION AND

STRUCTURE OF FUZZY

COGNITIVE MAPS

Methodological background of proposed approach is

based on a fuzzy cognitive maps as an extension of

knowledge maps first proposed by Robert Axelrod, a

political scientist, in 1976 (Axelrod, 1976). They

were used to present social scientific knowledge. A

fuzzy cognitive map is presented in the form of a

directed graph which can be represented in the

following manner (Froelich and Juszczuk, 2009):

<N,w>

(1)

where:

N = [N

,…N

]

- map factor values related to each

other by means of dependencies,

w = {w

}- connection weights assigned to the edges

between nodes x expressed in the form of relation

matrices, where are w

are numbers from the

interval [-1,1]; i,j = 1,….n, n- a number of factors.

Every edge w

is related to a given node N

and

has an attributed value. The value demonstrates a

kind of relations between factors. If an edge of a

node N

to a node N

has a value of > 0, it means a

positive influence of a factor A on a factor B. If an

edge coming from the factor B in the direction of A

has a negative value, it means that the factor B has a

negative influence on the factor A. When a value of

the edge equals 0, there is no mutual factor

influence. One needs to point out that the weight of

the edge w

≠ w

. A disadvantage of cognitive maps

was a presentation of relations between factors. The

presentation showed only a kind of connection.

Kosko suggested a change of a method for

determining node connection force (Kosko, 1986)

(Kosko and Postma, 1988). Instead of using marks

only, each edge had an assigned number which

determined the level of connection between

examined factors. Presented values were in the range

of [-1,1]. Consequently, the relations between

the factors could be described by means of fuzzy

terms, such as weak, medium or strong

(Kosko, 1986). The factor value depends

on determining map dynamics with a formula:







1













 x







∗w









(2)

where: i,j - factor numbers (i,j = 1, …n); n-

the number of factors; f - threshold number; t -

discreet time, x

- a value of i-th factor ; w

– a value

of edges between a factor x

and a factor x

(Froelich

and Juszczuk, 2009). The construction of a fuzzy

cognitive map is based, to a large extent, on input

date. This methodology uses the knowledge

of indicated subjects to represent their experiences

and behaviour by means of a map. The indicated

way of gathering information is subjective,

therefore, it is necessary to collect possibly

the largest group of experts or to rely on research

which included a broad sample (Sobczak, 2007).

Therefore, in the first stage one needs to gather main

factors which have the most vital influence

on a analysed phenomenon. The factors are chosen

on the basis of the number occurrences. If a factor,

among many independent experts’ opinions, occurs

many times, it ought to be included in the model

(Sobczak, 2007). The next step is to indicate

connections between the selected factors.

The connections need to be indicated on the basis

of real mutual factor interaction. Determining edges

and their direction allows defining interaction force

between them and it is determined on the basis

of experts’ knowledge. Interaction force of a relation

with regard to C

can be described by means of

linguistic variables (Papageorgiou and Kontogianni,

2011). Having assigned linguistic values from a set

T to the edges of the map one can determine

a numerical value to every edge. The fuzzification

of the obtained dependences between the nodes

of the map improves mapping of real relations

between the elements of the researched environment.

3 MODEL ASSUMPTIONS AND

THE EYE TRACKING

EXPERIMENT

The universality of FCM is expressed in its wide

application in various areas. In the literature one can

find examples of its application in solving

engineering (Mohr, 1997), industrial (Stylios

and Groumpos, 2004), military (Kosko, 1988)

or economic problems (Tsadiras and Margaritis,

1999). Cognitive maps can also be adapted to solve

problems related with the optimization of an online

advertising message. The issue of optimization

is concerned with dynamic changes of content

of an advertising message and its multitude.

By applying FCM one can quickly define

the effectiveness of a given combination and choose

the most favourable one. Such an approach

minimizes the time spent on keeping online low-

efficiency advertisements. Furthermore, in order

to minimize human contribution like in surveys

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

468

in estimating the value of correlation between map

nodes, eye tracking research was conducted. Still,

the approach involves human commitment and his

or her influence on shaping research,

and the approach also assumes that subjective

opinions are excluded. For the sake of the study

experimental websites had been prepared. Every

website contained four elements. Three of them

presented a text with a headline. The content

concerned current events related to e-commerce.

The fourth element showed an advertisement

constructed on the basis of three elements: level

of persuasion based on the colour intensity (P), size

(S) and location (L) on a website. Each feature could

appear in three variants what gave in total 27

combinations of advertisements which are possible

to obtain. Elements of websites with advertisements

were displayed on a 19-inch monitor with

the display resolution of 1024 x 768 pixels to 16

participants. The functioning of the device was

explained to every participant before the experiment.

Every participant’s position was appropriately set

to minimize individual differences in head

placement. After setting a proper angle and

a distance, the calibration process took place. In the

calibration procedure, which was 15 seconds long,

the participant’s task was to observe 9 points

arranged on the screen. After a proper calibration,

the user was acquainted with the procedures during

the experiment. The task involved reading headlines

and first sentences of every paragraph. Every screen

element presenting a website was shown for 15

seconds and then automatically another page was

displayed. The measurement was conducted with the

60Hz sampling rate. In order to measure

participants’ visual activity on every page there were

allotted areas of interest (AOI). Used measured

factors include the number of viewers to the

marketing content (MV), percentage of total time

spent on the website with the focus on advertisement

(MTP), focus time on the marketing content (MT),

time to the first view of the advertising content after

the website is fully loaded (MFV), the number of

repeated visits to the advertising content (MRV) and

the total number of returning visitors (MRVN).

Results for selected variants are showed in the Table

On the basis of data received, a number

of statistical analyzes was performed in order

to identify links between factors of P, L, S. The first

step was a conducted analysis of the correlation

between test components and test results obtained.

At a significance level of 0.05, the strongest

correlations were observed between the number

and location of revisited advertisement (-0.49).

The other element proved to be poorly correlated.

Cluster analysis based on the method of Ward

with a measure of Euclidean distance was used.

Based on the Mojeny rule, it is suggested that level

of division occurs at stage 2. Based on the resulting

graph, it can be determined by ad groups, which have

received similar results as the number of users who

noticed the ad. The best were advertising the group

in which repeated factor was the size of the ad (S)

at level 2, i.e. 250x200. To find out whether

a number is linked to a return visit with elements

of advertising, the obtained data was subjected

to analysis of main factors variance (ANOVA).

The number of ad visits was affected by its size

(F (2,20) = 4.32, p = 0.028). The analysis showed

that the ads, which are medium in size have a greater

impact on the return visit than ads with larger

dimensions. Their effectiveness is about 23% higher.

Based on these results, important factors proved

to be the location (p = 0.003) and ad size (p = 0.014).

The intensity of the colour turned out not

to be significant (p = 0.313).

4 THE FORM OF THE FCM

MODEL FOR CHOOSING

ADVERTISING CONTENT

The data obtained from measurements constitute N

factors containing information about characteristics

Table 1: The values from eye tracking for selected variants.

Factors

MV MFV MT MTP MRVN MRV

1-1-1 7 5.2 0.56 3.76 5 1.4

2-3-2 9 7.27 0.46 3.09 1 1

2-2-2 9 7.39 0.67 4.49 5 1.8

1-1-3 8 3.2 0.68 4.51 6 2.7

3-3-1 5 8.14 0.41 2.71 1 1

3-2-1 2 6.75 0.06 0.37 1 1

2-1-2 6 8.68 0.21 1.4 2 1.5

2-3-3 5 8.48 0.66 4.37 2 1.5

1-3-2 5 4.71 1.21 8.06 3 1.7

1-3-3 8 5.72 0.2 1.32 4 1.3

The Use of a Fuzzy Cognitive Maps and Eye Tracking in Exploitation of Online Advertising Resources

469

of an advertisement and its effectiveness.

Independent variables, which can be controlled

in a direct way, are the size, colour and location

of the advertisement. Other information obtained

on the basis of the research by means of the device

are seen as dependent variables whose value

depends on a user’s reaction. Here is the list of all

variables (nodes): N

- advertisement size (S); N

location of the advertisement on a website (L); N

persuasion based on colour intensity (P); N

–

the number of visits; N

- time after which

the advertisement is noticed for the first time (s);

- total time in which the advertisement caught

attention (s); N

- percentage share time

of advertisement visits in relation to other content

on the website (%); N

- the number of return visits.

Figure 1: Diagram of a fuzzy cognitive map for the issue

of advertisement content and localisation selection.

The initialisation of map parameters in a relation

matrix was constructed on the basis of the analysis

of parameters obtained from the eye tracking

research and is showed in Fig. 1. The calculated

correlations were used to create connections

between map nodes w

. The matrix based

on the obtained values is presented in the Table 2.

The zero value means that given nodes do not

influence significantly each other.

On the basis of charts of surface adjustment

between individual advertisement elements one can

observe the changeability of effectiveness

of individual combinations with relation to intensity

of each feature. Fig. 2 shows the relationship

between two most essential factors. The surface

chart displays that most effective is the combination

of location on level L=1 and of size on level S=2.

Figure 3 presents the relation between the location

and colour intensity. It turns out that the best results

are achieved for an advertisement placed

on the highest place on a website and for colour

on lever P=2. The location change is responsible

for a drop of users’ return visits. The diversity

of a mutual influence wielded by the size and colour

intensity turned out to be irrelevant.

Figure 2: Surface adjustment Size – Location.

Figure 3: Surface adjustment Colour – Location.

The prepared cognitive map also allows conducting,

in a short period of time, a large number of

simulations for different values of decision

variables. When the defining values

of advertisement elements are introduced

to the model, it is possible to verify the results

of effectiveness which a given combination

of elements is able to achieve. Also, the proposed

model can be used to predict effectiveness results.

When introducing a demanded value to a given node

from N

to N

, one can define which values adopt

independent nodes. This solution enables the

dynamic selection of advertisement content

depending on the desired results. This solution

Table 2: Relation matrix of the discussed map.

0 0 0 -0.27 0 0 0 -0.49

0 0 0 0 0.27 0 0 0

0 0 0 0.27 0 0 0 0

-0.27 0 0.27 0 0 0.30 0.30 0.62

0 0.27 0 0 0 0 0 -0.35

0 0 0 0.30 0 0 0.99 0.40

0 0 0 0.30 0 0.99 0 0.40

-0.49 0 0 0.62 -0.35 0.40 0.40 0

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

470

seems to be correct given the fact that in advertising

campaigns the demanded effect can be diversified.

Figures 4 - 5 represent selected part of possible

scenarios. Values in highlighted nodes refer

to elements constructing advertising content (Fig. 4 -

5) or a predicted value (Fig. 6 - 7).

Figure 4: Effectiveness results for settings L=2, S=1, P=1.

Figure 5: Effectiveness results for settings L=3, S=2, P=3.

Figure 6: Forecasted values of independent nodes for an

expected value of a node N4 ≥ 8.0.

Figure 7: Forecasted values of independent nodes for an

expected value of a node N8 ≥ 6.0.

Figures 4 and 5 present achievable effectiveness

results of nodes N

-N

. Generated results are based

in full on the values of nodes N

, N

and N

. In both

cases a given list of independent variables generates

a given domain of results. In cases presented

in Figures 6 and 7, FCM was used to predict

the settings of independent variables. The forecast

range of values in both cases is an inequality.

Because of that, the number of generated settings,

which fulfil this inequity, of advertisement content

is higher than one. Other calculations will generate

different element combinations in nodes N

-N

The conducted research also allowed defining

the influence of individual variables on each other.

This illustrates the complexity of the modelled

system of advertising content selection.

5 SUMMARY

The introduced modelling of an interactive message

is an example of using fuzzy knowledge maps

combined with the eye tracking. The presentation

of changeability of features constructing a message

as well as the evaluation of effectiveness of results

obtained in the model makes it possible to have

a wider range of decision support with relation

to deterministic solutions. The obtained results,

because of the possibility of evaluation of their

realization certainty in a changeable environment,

indicate that fuzzy cognitive maps can be employed

in the optimization of planning a transmission

of a certain type of advertisement on the Internet.

They can be applied in strategic planning when

looking for solutions which ensure the best use

of available resources and be the basis for

the evaluation of broadcasting resources from

a point of view of achieved results. In order to test

the universality of the model it is possible to adapt

a similarly constructed advertisement and to check

the correctness of the results generated by FCM.

Adding other nodes with independent values

to the constructed map would enable stronger

personalization of a generated message, what

in consequence would increase the level

of effectiveness.

REFERENCES

Axelrod R., 1976. Structure of Decision: The Cognitive

Maps of Political Elites, Princeton University Press,

Princeton, NJ.

Froelich W., Juszczuk P., 2009. Predictive Capabilities of

The Use of a Fuzzy Cognitive Maps and Eye Tracking in Exploitation of Online Advertising Resources

471

Adaptive and Evolutionary Fuzzy Cognitive Maps – A

Comparative Study. Intelligent System for Knowledge

Management (Studies in Computational Intelligence

Series), Vol. 252. Springer-Verlag, Berlin-Heidelberg,

pp. 153-174.

Hoffman D.L., Novak T.P, 1995. Marketing in

Hipermedia Computer-Mediated Environment:

Conceptual Foundations, Working Paper No. 1,

Vanderbilt University, Nashville.

Kosko B., 1986. Fuzzy cognitive maps, Internat. J. Man-

Mach. Studies 24, pp. 65-75.

Kosko B., 1988. Hidden patterns in combined and

adaptive knowledge networks, Internat. J. Approx.

Reason. 2, pp. 377-393.

Kotler Ph., Postma P., 1999. The New Marketing Era,

Marketing to the Imagination in a Technology World,

McGraw-Hill, New York.

Mohr S., 1997. The use and interpretation of fuzzy

cognitive maps, Master’s Project, Rensselaer

Polytechnic Institute (1997).

Papageorgiou E., Kontogianni A, 2011. Using Fuzzy

Cognitive Mapping in environmental decision making

and management: a methodological primer and an

application. S.S. Young, S.E. Silvern (Eds.),

International Perspectives on Global Environmental

Change, InTech Open Access Publisher, pp. 427-450.

Rust T., 1989.Advertising Media Models, Lexington

Books.

Sobczak A., 2007. Zastosowanie rozmytych map

kognitywnych w planowaniu rozwoju zorientowanej na

usługi architektury systemów informatycznych,

Zarządzanie rozwojem organizacji, Wydawnictwo

Politechniki Łódzkiej.

Stylios C., Groumpos P, 2004. Modelling complex systems

using fuzzy cognitive maps, IEEE Trans. Systems

Man, Cybern. Part A: Systems Humans 34 (1), pp.

155-162.

Tsadiras A., Margaritis K., 1999. An experimental study

ofthe dynamics of the certainty neuron fuzzy cognitive

maps, Neurocomputing 24, pp. 95-116.

Goldfarb A., Tucker C. E., 2011. Online display

advertising: targeting and obtrusiveness, Marketing

Science 30(3), pp. 389-404.

Theocharous G., Thomas P.S, Ghavamzadeh M., 2015. Ad

Recommendation Systems for Life-Time Value

Optimization. In: Proceedings of the 24th International

Conference on World Wide Web, Republic and

Canton of Geneva, Switzerland, pp. 1305-1310.

Wan-Shiou Y., Jia-Ben D., Hung-Chi Ch., Hsing-Tzu L.,

2006. Mining Social Networks for Targeted

Advertising, Proceedings of the 39th Annual Hawaii

International Conference on System Sciences

(HICSS'06), Track 6, vol. 6, pp. 137-147.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

472