Emotion Recognition through Keystroke Dynamics on Touchscreen

Keyboards

Matthias Trojahn

, Florian Arndt

, Markus Weinmann

and Frank Ortmeier

Volkswagen AG, Wolfsburg, Germany

TU Braunschweig, Institute of Business Information Systems, Chair of Information Management, Braunschweig, Germany

Otto-von-Guericke University of Magdeburg, Computer Systems in Engineering, Magdeburg, Germany

Keywords:

Human-Computer Interaction, Emotion Recognition, Keystroke Dynamics, Touchscreen, Smartphone.

Abstract:

Automatic emotion recognition through computers offers a lot of advantages, as the interaction between hu-

man and computers can be improved. For example, it is possible to be responsive to anger or frustration of

customers automatically while working with a webpage. Mouse cursor movements and keystroke dynamics

were already used and examined for such a recognition on conventional keyboards.

The aim of this work is to investigate keystroke dynamics on touchscreen keyboards which gets a cumulative

relevance through the increasingly further circulation of smartphones and tablets. Furthermore, it is possible

to record additional information like pressure and size of keystrokes. This could increase the recognition rate

for emotions. In order to record the keystroke dynamics, an application and keyboard layout for Android OS

were developed. In addition, hypotheses were established on the basis of Yerkes-Dodson-Law and Flow the-

ory and besides, a study with 152 test persons for the data collection was implemented. Subsequently, a data

evaluation with the SPSS software was accomplished. Most of the hypotheses were conﬁrmed and the results

of the study show that emotions can be explained by the keystroke dynamics and recognized in this way.

1 INTRODUCTION

The human-computer interaction (HCI) is getting a

rising relevance in the last few years because of the

cumulative circulation of computers in our everyday

life. Nowadays, almost everybody interacts with com-

puters in different ways. For example the use of an

elevator which is controlled by a computer or by the

use of a navigation system which can be designated

as a computer. One way to improve an interaction is

the automatic recognition and reaction of emotions.

This is relevant for the human-human interaction and

the human-computer interaction. The recognition en-

ables the automatic reaction for example to frustration

and anger, which can occur during a customer pro-

cess or a visit on a webpage. This improvement of

working with computers can result in positive emo-

tion which, furthermore, can increase the productiv-

ity and also the health of the working person (Cohen

et al., 2003).

The subject of emotion recognition has already

been examined with mouse cursor movements and

keystroke dynamics on conventional keyboards. Be-

cause of the rising circulation of smartphones, tablets

and general touchscreens in the last few years the aim

of this work is to consider the keystroke dynamics on

touchscreen keyboards for possible emotion recogni-

tion. This kind of keystroke dynamics has so far only

been used for biometric authentication analysis (Bu-

choux and Clarke, 2008).

Within the framework of this subject an applica-

tion and a special keyboard layout for the Android

operating system were developed. For the theoreti-

cal background the Yerkes-Dodson-Law (Yerkes and

Dodson, 1908) and the Flow theory (Csikszentmiha-

lyi, 1975) were consulted to control the hypotheses.

Subsequently, a study of 152 test persons as a data

collection was implemented and an evaluation with

the SPSS software was accomplished.

Section 2 will give a short theoretical background

about emotion and keystroke dynamics in combina-

tion with two theories and our hypotheses. Then, we

will introduce the empirical study in Section 3. In the

next section we will describe the results of the study

which we will discuss in Section 5 with the existing

limitations and implications.

Trojahn M., Arndt F., Weinmann M. and Ortmeier F..

Emotion Recognition through Keystroke Dynamics on Touchscreen Keyboards.

DOI: 10.5220/0004415500310037

In Proceedings of the 15th International Conference on Enterprise Information Systems (ICEIS-2013), pages 31-37

ISBN: 978-989-8565-61-7

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2 THEORETICAL BACKGROUND

2.1 Emotion, Stress, Keystroke

Dynamics

Emotions are an important function of the human

body to react on dangerous or extraordinary situa-

tions. Because the process of decision making of hu-

man beings is very slow, emotions can help us to re-

spond to the current circumstances in an efﬁcient and

fast way (Maehr, 2008). Negative emotions warn us if

a given aim cannot be reached and positive emotions

calm down if the situation is under control. Further-

more, emotions often have a speciﬁc cause and usu-

ally are an intense experience of short duration (Zim-

mermann et al., 2003).

To classify the emotions, a theory with three di-

mensions has been developed by Mehrabian (1970),

who divides all emotions into three categories

(Mehrabian, 1970). The three dimensions are arousal,

valence and control. The most variance exists in the

ﬁrst two dimensions. Therefore, the third dimension

is often unnecessary and will be also not recorded in

this study (Bradley and Lang, 1994).

Stress is often treated in relation with emotions

(Lazarus, 2006). Stress is a synonym for pressure or

tension. It describes the reaction of exterior stimuli,

which enable the accomplishment of certain exercises

and the resulting psychic and physical burden (Selye,

1936). These exterior stimuli can be noise, injuries,

coldness or excessive demand. In general, two differ-

ent types of stress exist: Eustress is the positive expe-

rienced activation of the organism and distress is the

burdensome and harmful stress (Selye, 1975). Nowa-

days, the latter meaning is often used for stress.

Keystroke dynamics is a biometrical attribute of

every human being like ﬁngerprint, retinal scan or

voice recognition (Amberg et al., 2003). Unlike a

password which a person knows or an identiﬁca-

tion card which a person owns, a biometrical fea-

ture is a characteristic of a person herself (Buchoux

and Clarke, 2008). The keystroke dynamics can be

matched with the handwritten signature (Joyce and

Gupta, 1990). In this case, the keyboard input is

monitored in order to identify a pattern of tip rhythm

(Monrose and Rubin, 1997). Keystroke dynamics for

a touchscreen display on smartphones was already

used by Trojahn et al. (Trojahn and Ortmeier, 2012).

In addition to time differences, pressure and size dur-

ing typing were recorded and used for authentication.

2.2 The Yerkes-Dodson-Law and the

Flow-theory

The Yerkes-Dodson-Law describes the context be-

tween the productivity and the activity respectively

the arousal of a person (Yerkes and Dodson, 1908). If

the arousal or rather stress increases the productivity

increases too. After a peak the productivity decreases.

The peak depends on the respective person and the

difﬁculty of the task. After the peak the stress is too

high and the productivity decreases. The person feels

negative stress. This is an example of positive and

negative stress which was developed 1908 by Robert

M. Yerkes and John D. Dodson.

The context between the challenge and the skills

of a person is described by the Flow Theory which

was invented by Mihaly Csikszentmihalyi in 1975

(Csikszentmihalyi, 1975). If the skills correspond to

the current challenge of a task, a state of Flow and

positive emotions come up. At this state the person is

completely concentrated on his/her work and he/she

does not recognize the environment around him/her.

To get in this state some conditions have to be ful-

ﬁlled. The goals of the task must be clear for the per-

son and the person must have conﬁdence in being able

to fulﬁll the task.

2.3 Hypotheses

First of all, to study the inﬂuence of the emotions

on the keystroke dynamics, hypotheses were derived

from the theories. A time limit was initiated to evoke

stress (Lazarus et al., 1952). According to the Yerkes-

Dodson-Law the productivity rises through increas-

ing activation or increasing stress. After exceeding a

vertex, the productivity sinks again. These levels are

mentioned as positive and negative stress levels.

The increased activation causes a ﬂow state of the

participants because of the increased productivity and

the skills which are on this way adapted. This ﬂow

state is accompanied by positive emotions. Through

further reduction of the time limit to complete the typ-

ing of the text, the stress level rises. The arise of neg-

ative emotions results from the interrupted ﬂow state.

Stress is created by the reduction of the time limit

which causes emotions. This represents the ﬁrst two

hypotheses.

H1: The closer the time limit, the higher the ex-

citement.

H2: A signiﬁcant difference of the valence can be

observed between the groups.

The inﬂuence of the emotions on the keystroke dy-

namics was supposed to be examined. Former studies

showed that the typing speed decreases when negative

ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems

emotions are perceived (Khanna and M.Sasikumar,

2010). Nevertheless, the error rate rises by the occur-

rence of negative emotions (Alepis and Virvou, 2006).

The same effect can be observed at the pressure of key

presses. This should also increase by the confronta-

tion of negative emotions (Alepis and Virvou, 2006).

Further hypotheses can be derived from that.

H3: The slower the typing speed, the more nega-

tively emotions are perceived.

H4: The higher the typing error rate, the more

negatively emotions are perceived.

H5: The higher the typing pressure, the more neg-

atively emotions are perceived.

These ﬁve hypotheses were supposed to be exam-

ined within the framework of the experiment for the

keystroke dynamics on touchscreen keyboards.

3 EMPIRICAL STUDY

In order to examine the keystroke dynamics an exper-

iment was developed. Its design, the participants, the

measured variables and the model speciﬁcation are

described in the next few sections.

3.1 Study Object and Study Task

In order to investigate the keystroke dynamics two

identical Samsung Galaxy Nexus were used. The test

persons were asked to enter some descriptive data like

age, gender, experience and education level. After

that, they were asked to enter a text which was pre-

sented to them in printed version. We tried to select a

neutral text which does not cause any emotions. Ac-

cording to Wagner et al. (2001) we took a text about

a bronze statue (Wagner et al., 2001). For the pro-

cess of typing they got randomly one of three differ-

ent time limits of either “no time limit”, “5 minutes”

or ”3.5 minutes“. The time limits represent the differ-

ent stress levels. Therefore, the time limits were cal-

culated in a way that the task of typing the text could

barely or not be mastered. After typing the text, they

were asked to ﬁll out an emotion questionnaire which

was also handed to the test person in a printed ver-

sion. This was created to record the emotions and the

arousal which occurred during typing the text under

pressure with a time limit.

We developed the software keyboard and the study

application ourselves. They record the keystroke dy-

namics data and the descriptive data in an output

ﬁle which is imported by another application into a

MySQL database. By the use of export software, the

data were converted and some variables were calcu-

lated to evaluate it with the SPSS software.

3.2 Study Design and Subject

Participants

The design of a randomized laboratory experiment

was selected. Gender, age and experience in the han-

dling of a touchscreen can be described as random

variables. The study was accomplished during the pe-

riod of the 29-05-2012 to 08-07-2012. Overall, 152

test persons were consulted and the study lasted 15-20

minutes. The average age of all test persons was 31.14

years and the percentage of women was 23.68 %. We

tried to get a regular distribution of the test persons

over the three different time limits. Furthermore, we

tried to get an equal number of male with and without

experience and female with and without experience

in every time limit. Figure 1 shows the distribution of

the three time limits and the four different groups of

persons subdivided in experience and gender.

Figure 1: Distribution of the test persons subdivided into

time limit, experience, and gender.

The diagram shows that almost an equal number

of test persons in every time limit was reached (50 at

no time limit, 50 at 5 minutes, 52 at 3.5 minutes). The

number of male persons with experience is the largest.

This can be traced back to the fact that most test per-

sons are employees of an IT department or students of

a technical university. The smallest part is the number

of female test persons without experience.

3.3 Measure Variables

Emotion was chosen to be the dependent variable. It

was recorded by the use of the emotion questionnaire,

digitalized and afterwards also saved in the database.

The arousal and the valence were measured and two

different kinds of questionnaires (SAM and PANAS)

were used. In this way the results could be compared

and reviewed.

The keystroke dynamics variables were inserted

as independent variables. In relation to the hypothe-

ses the typing speed, error rate and pressure of the

key presses were considered. These are calculated by

EmotionRecognitionthroughKeystrokeDynamicsonTouchscreenKeyboards

Table 1: Results of the correlation analysis.

SAM val. SAM ar. speed error rate pressure gender age time edua.

SAM va-

lence

SAM arousal 0.308*** 1

speed -0.131*** -0.020 1

error rate 0.075*** 0.056*** -0.294*** 1

pressure -0.046*** -0.018 0.155*** 0.023* 1

gender 0.046*** -0.168*** -0.043*** -0.035*** 0.122*** 1

age 0.047*** -0.281*** -0.218*** -0.019 -0.121*** 0.282*** 1

time limit 0.003 0.205*** -0.008 -0.65*** 0.042*** -0.002 -0.021* 1

education

level

-0.222*** -0.026** 0.003*** -0.046*** 0.035*** -0.058*** -0.1*** 0.071*** 1

* p <0.10; ** p <0.05; *** p <0.01; 2-sided test.

the export of the data from the database into a CSV

ﬁle for the evaluation with the SPSS software. All of

the variables refer to single words. That means that

word averages of the variables are formed for all test

persons. The calculation of the typing speed is mea-

sured by the timestamp and the number of key presses

of one word. The error rate describes the number of

presses of the delete key. Because of this, the error

rate describes the number of the conscious failure. To

use the pressure an average value of all keystrokes for

every word was calculated.

As control variables some different descriptive in-

formation of the test persons are applied. Age, gender,

education level and the assigned time limit of the test

persons were used for the evaluation. All information,

except of the time limit, are entered over the applica-

tion by the test persons themselves. The time limit is

assigned randomly and entered by the examiner after

the experiment for every study participant.

3.4 Model Speciﬁcation

For the test of the hypotheses 3 to 5 we developed

a linear regression model. It is shown in Formula 1

below.

emotions =α + β

× speed + β

× f ailure count

+ β

× pressure + controls + ε

(1)

The control variables are age, gender, education

level and time limit of the test persons. In order to

be able to conﬁrm the hypotheses 3 the p-value of the

must be signiﬁcantly and smaller zero. For the hy-

potheses 4 respectively 5 the p-value of the β

respec-

tively β

has to be signiﬁcantly and greater zero.

In the evaluation four models with different num-

bers of control variables were deﬁned. The ﬁrst model

has no control variables except of the dependent and

independent variables. For the second model age and

gender of the test persons are included. In the third

model the time limit and in the fourth model also the

education level is added. In this way the inﬂuence of

the individual control variables on the model can be

shown.

4 RESULTS

In the beginning of the evaluation a correlation anal-

ysis was used. Therefore, the coherences of the vari-

ables can be shown. Table 1 offers the results of the

correlation analysis for all considered variables.

In Table 1 the correlation coefﬁcients between ev-

ery variable are mapped. The asterisks indicate the

signiﬁcance level and are therefore called signiﬁcance

asterisks. A high number of asterisks entail a high sig-

niﬁcance level. In the ﬁrst column it can be observed

that between the SAM valence variable and nearly all

other variables the signiﬁcance level is high. That

means that a signiﬁcant connection between these

variables exists. Another interesting connection is be-

tween the SAM arousal and the time limit. That in-

dicates a connection between the time limit and the

caused stress.

To prove the hypotheses 1 and 2 we also imple-

mented an ANOVA analysis for an average compari-

son. This calculation compares the averages and the

variances between partial samples. For this study, the

groups are divided into time limits. In this way, we

get three different random samples. The results of the

average comparison are shown in Table 2.

In the second row of Table 2 the numbers of con-

sidered objects (N) are given. The objects represent

the entered words of all test persons. This number

decreases by a smaller time limit because the test per-

sons do not accomplish all words in the given time

limit. It also shows that the time limit is so straight-

ened that a complete input of the given text is not pos-

sible. In the third and fourth row the mean values and

ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems

Table 2: Results of the ANOVA analysis.

sample feature no time limit 5 minutes 3,5 minutes F-value

N 2158 2124 1894

Arousal 3.630 (1.858) 4.410 (2.018) 4.660 (2.138) 147.349***

Valence 4.480 (1.656) 4.213 (1.547) 4.503 (1.843) 19.171***

* p <0.10; ** p <0.05; *** p <0.01; 2-sided test.

the variance values of the both variables in the three

groups are illustrated. In consideration of the values

and the F-values in the last column it is to be rec-

ognized that for the valance and the arousal variable

signiﬁcant differences exist between the groups. If no

differences exist the F-value is nearly zero. It can be

observed that the F-values of the variables are not zero

and signiﬁcant. That means the arousal and the va-

lence is signiﬁcantly different between the groups. In

addition, we found out that the arousal rises at closer

time limit. Hypotheses 1 and 2 can be conﬁrmed with

this.

The Figure 2 shows a diagram that represents the

averages values of both variables over the time limits.

Figure 2: Diagram of the averages values of the arousal and

valence variable over the three time limit groups.

By means of the diagram the differences and the

rise of the arousal variable can be recognized.

In order to ﬁnd the type of the connection between

the variables a linear regression analysis was made.

The valence is used as dependent variable which is

supposed to be explained by the keystroke dynamics

variables. Table 3 shows the results of the regression

analysis. The four different models with the different

number of control variables can be recognized.

The last row shows the number of the considered

objects again. In the second, third and fourth row the

keystroke dynamics variables are mapped. The con-

trol variables can be found from the ﬁfth to the eighth

row. In all four models the results of the keystroke dy-

namics variables are signiﬁcant which is displayed by

the signiﬁcant asterisks. The coefﬁcient of determi-

nation R

in the ﬁrst three models of 2 % is relatively

low. By addition of the education level variable as a

control variable the coefﬁcient rises to 7 %. The ad-

justed R

value is related to the number of variables

that are observed and tries to correct the rising of the

value by just adding more variables to the model.

Low coefﬁcient of determination are to be considered

critically, but according to Davis (1989) (Davis, 1989)

or Moon and Kim (2001) (Moon and Kim, 2001) low

values in the behavior-oriented social research are ab-

solutely common.

For the conﬁrmation of the third hypothesis the

regression coefﬁcient of the keystroke speed needs to

be signiﬁcant and smaller than zero. For the hypoth-

esis 4 the coefﬁcient of the error rate must be signiﬁ-

cant and greater than zero. The results show that these

conditions are given in all four models. Thus the hy-

potheses can be approved. The speed decreases and

the error rate increases by the perception of negative

emotions. For the last hypothesis the regression co-

efﬁcient of the keystroke pressure must be signiﬁcant

and also greater than zero. On the basis of the results

of the fourth row it is shown that the coefﬁcient is sig-

niﬁcant in all four models but the coefﬁcient is smaller

than zero. Thus the hypothesis must be rejected.

Although the hypothesis was developed by means

of the literature it could not be conﬁrmed. Therefore,

this topic will require further research in future stud-

ies.

5 CONCLUSIONS

5.1 Summary

Through the evaluation of the collected data with

the SPSS software most of the developed hypothe-

ses could be approved. Stress and emotions can be

caused by the commitment of time limits. These

evoked emotions can be explained by the keystroke

dynamics variables keystroke speed and error rate.

Thus, the ﬁrst four hypotheses could be conﬁrmed.

The keystroke pressure becomes signiﬁcant smaller

by the perception of negative emotions. Therefore,

the last hypothesis has to be rejected and more stud-

ies are required in order to keep on examining this

phenomenon.

EmotionRecognitionthroughKeystrokeDynamicsonTouchscreenKeyboards

Table 3: Results of the regression analysis.

-1 -2 -3 -4

keystroke speed -0.106*** -0.102*** -0.102*** -0.111***

failure count 0.061*** 0.066*** 0.067*** 0.047**

keystroke pressure -0.545** -0.642*** -0.648*** -0.448*

gender 0.181*** 0.182*** 0.169***

age 0.001 0.001 -0.003

time limit 0.013 0.052**

education level -0.285***

constant 4.982 4.848 4.837 5.523

0.019 0.022 0.022 0.070

adjusted R

0.019 0.021 0.021 0.069

N 6221 6221 6221 6221

* p <0.10; ** p <0.05; *** p <0.01; 1-sided test.

5.2 Limitations

Like in every other study some limitations occur in

this study which is supposed to be named and ex-

plained in the following. First of all, it must be ex-

plained that the effect of the exercise itself cannot

be controlled because only one exercise was consid-

ered. Furthermore, the study was implemented in

a controlled setting. This means that not all exte-

rior inﬂuences were regarded. It is possible that the

test persons had already negative emotions or were

stressed before the experiment started. This might

have distorted the results of the study, respectively the

recorded data of the emotion questionnaire. Further-

more, the inﬂuence of the hardware setting should be

named. The study was implemented with a smart-

phone with and Android operation system. Test per-

sons who have already used an Android smartphone

before have more experience in the handling which

could also distort the results in this way. Through the

development of a self-constructed keyboard this ef-

fect should be avoided. Nevertheless, this cannot be

completely assured.

The small percentage of female participants and

the unevenly distribution of the education level are

further limitations of the study. Also the rate of the

persons who accomplished the complete text within

the 3.5 minutes can have an inﬂuence on the results.

Because they reached the aim for entering the text

in the pre-set time limit, they were most probably

stressed, but in a positive way. At the closest time

limit it was planned that negative emotions are caused

by non-attaining. However, the part of the persons

who managed it in spite of the close time limit was

only near 29 %.

Moreover, also the small coefﬁcient of determi-

nation must be named as a limitation. As already

mentioned, small values are normal in the behavior-

oriented social research (Davis, 1989; Moon and Kim,

2001). Nevertheless, there should be more research

into this topic in future studies. In addition, the re-

sults for the control variables should be mentioned.

The regression coefﬁcients for example of the gender

variable are signiﬁcant in all four models. This im-

plies that it has a signiﬁcant inﬂuence on the depen-

dent variable and would have to be considered, there-

fore, actually as an independent variable.

With respect to a commercial use even the data

privacy has to be considered. The self-developed soft-

ware is in principle a key logger which can be used to

save passwords or other sensitive information. Solu-

tions for that must be clariﬁed in the case of an intro-

duction. An option would be to recognize automati-

cally the input of passwords and to stop the saving for

this input.

Last but not least we should refer to the not con-

ﬁrmed hypothesis of the keystroke pressure. By the

feeling of negative emotions the pressure is smaller,

not greater. This cannot be occupied by means of the

currently literature, so more studies are required in

this section. Because the pressure can be recorded

on conventional keyboards exclusively with supple-

mentary hardware, the number of studies is up to now

small in this range. In the coming years this will prob-

ably change because through the increasingly further

circulation of touchscreens a lot of new possibilities

will occur.

5.3 Implications

The study and the results show that the described

Yerkes-Dodson-Law and the Flow theory could be

validated. The evaluation proves that positive stress

exists and that it can be initiated. Furthermore, the

productivity increases by the perception of positive

stress. This represents important results for the be-

havior and stress research in the work environment.

The main intention of this study, to investigate the

automatic recognition of emotions through keystroke

dynamics on touchscreen-keyboards, was processed

ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems

and successfully completed. Emotions can be ex-

plained by the keystroke dynamics. This opens new

possibilities for marketing departments and interface

designers who can deal with customers purposefully.

With the aid of these ﬁndings it is possible to develop

systems which give supports automatically in the case

of recognition of e.g. frustration. In the e-commerce

environment this can be enormously helpful and rep-

resent a competitive advantage.

Also in the domain of the authentication with the

use of the keystroke dynamics the results should ﬁnd

attention because it was shown that the keystroke dy-

namics is strongly inﬂuenced by emotions. This in-

ﬂuence must be considered in relation to the develop-

ment of the algorithm and during the training of the

algorithm for an authentication.

REFERENCES

Alepis, E. and Virvou, M. (2006). Emotional intelli-

gence: constructing user stereotypes for affective bi-

modal interaction. In Proceedings of the 10th in-

ternational conference on Knowledge-Based Intelli-

gent Information and Engineering Systems - Volume

Part I, KES’06, pages 435–442, Berlin, Heidelberg.

Springer-Verlag.

Amberg, M., Fischer, S., and R

oßler, J. (2003).

Biometrische Verfahren - Studie zum State of

the Art. Technical report, Friedrich-Alexander-

Universit

at Erlangen-N

urnberg.

Bradley, M. M. and Lang, P. J. (1994). Measuring emotion:

The self-assessment manikin and the semantic differ-

ential. Journal of Behavior Therapy and Experimental

Psychiatry, 25(1):49 – 59.

Buchoux, A. and Clarke, N. L. (2008). Deployment of

keystroke analysis on a smartphone. In Proceedings

of the 6th Australian Information Security & Manage-

ment Conference.

Cohen, S., Doyle, W. J., Turner, R. B., Alper, C. M., and

Skoner, D. P. (2003). Emotional style and suscepti-

bility to the common cold. Psychosomatic Medicine,

65(4):652–657.

Csikszentmihalyi, M. (1975). Play and intrinsic rewards.

Journal of Humanistic Psychology, 15(3):41–63.

Davis, F. D. (1989). Perceived usefulness, perceived ease of

use, and user acceptance of information technology.

MIS Quarterly, (Sept.):319–339.

Joyce, R. and Gupta, G. (1990). Identity authentica-

tion based on keystroke latencies. Commun. ACM,

33(2):168–176.

Khanna, P. and M.Sasikumar (2010). Recognising emotions

from keyboard stroke pattern. International Journal of

Computer Applications, 11(9):1–5.

Lazarus, R. S., Deese, J., and Osler, S. F. (1952). The effects

of psychological stress upon performance. Psycholog-

ical Bulletin 49, 4:293–317.

Lazarus, R. S. P. (2006). Stress and Emotion: A New Syn-

thesis. Springer Series. Springer Publishing Company.

Maehr, W. (2008). eMotion: Estimation of User’s Emo-

tional State by Mouse Motions. VDM Verlag,

Saarbr

ucken, Germany.

Mehrabian, A. (1970). A semantic space for nonverbal be-

havior. Journal of Consulting and Clinical Psychol-

ogy, 35 (2):248–257.

Monrose, F. and Rubin, A. (1997). Authentication via

keystroke dynamics. In Proceedings of the 4th ACM

conference on Computer and communications secu-

rity, CCS ’97, pages 48–56, New York, NY, USA.

ACM.

Moon, J.-W. and Kim, Y.-G. (2001). Extending the tam for

a world-wide-web context. Information and Manage-

ment, 38(4):217 – 230.

Selye, H. (1936). A syndrome produced by diverse nocuous

agents. Nature, 138:32.

Selye, H. (1975). Stress and distress. Comprehensive Ther-

apy, 1 (8):9–13.

Trojahn, M. and Ortmeier, F. (2012). Biometric authenti-

cation through a virtual keyboard for smartphones. In

International Journal of Computer Science & Infor-

mation Technology (IJCSIT).

Wagner, U., Gais, S., and Born, J. (2001). Emotional mem-

ory formation is enhanced across sleep intervals with

high amounts of rapid eye movement sleep. Learning

& Memory, 8(2):112–119.

Yerkes, R. M. and Dodson, J. D. (1908). The relation

of strength of stimulus to rapidity of habit-formation.

Journal of Comparative Neurology and Psychology,

18(5):459–482.

Zimmermann, P., Guttormsen, S., Danuser, B., and Gomez,

P. (2003). Affective computing–a rationale for mea-

suring mood with mouse and keyboard. International

Journal of Occupational Safety and Ergonomics.

EmotionRecognitionthroughKeystrokeDynamicsonTouchscreenKeyboards