TypeVis: Visualization of Keystrokes and Typing Patterns based on Time

Series Analysis

Kinga Enyedi and Roland Kunkli

Faculty of Informatics, University of Debrecen, Debrecen, Hungary

Keywords:

Typing Pattern, Time Series, Typing Style, Keystrokes.

Abstract:

In recent decades, people spend more and more time in front of their computers; therefore, instead of writing

on a paper, it is more common to type on a keyboard. In the case of handwriting, there are personal traits

implicitly left with the written text since people do not write the same way. Since how one writes on a paper

using a writing tool results in a different outlook of the written text for each person, one may question whether

the process of typing on a keyboard involves any speciﬁc personality traits or not. It can be proven that several

particular properties exist in the case of keyboard usage too. One of the most typical properties is based on

differences in time with respect to pressing and releasing keys on the keyboard. In this paper, we present a

visualization method for representing the typing styles of people, based on differences in when they press and

release each key on the keyboard over time. Using time differences of keystrokes for visualization purposes

allows to create a visible image about an individual’s personal typing habits.

1 INTRODUCTION

In our digitized world, handwriting is a little less po-

pular than before. With handwriting, personal traits

like the shape of the letters or the space among them

can clearly be seen in the written text. However, when

typing on a keyboard, we cannot see such traits di-

rectly in the typed text, while it is straightforward that

the way how people type is not the same at all: many

factors that determine the typing style is different for

each person (Baker and Redfern, 2007), (Longi et al.,

2015). For example, force and pressure used to press

a key, or the velocity of the key after pressing it are

usually different, just like the number of ﬁngers used

to type or the position where eyes look while typing.

It is clear that such properties can be speciﬁc for indi-

viduals.

Based on the way how these traits can be measu-

red, we can categorize them into two groups: there are

properties just like the amount of pressure on the keys

that would require special hardware to measure, and

there are ones that can be measured based on time,

that means they can be captured by using an ordinary

keyboard. In this work, we will use time differences

based typing properties to distinguish and visualize

individuals’ typing styles.

Many studies exist that have already used time-

based personal typing characteristics for speciﬁc pur-

poses proving the effectiveness of using an indivi-

dual’s typing style as the base of user authentication

and identiﬁcation (Karnan et al., 2011). Using key-

stroke dynamics, there is a personal identity veriﬁca-

tion method presented with a false alarm rate of less

than 5% (Gunetti and Picardi, 2005) and less than 4%

(Bergadano et al., 2002). In another paper, time se-

ries analytics was used as well in connection with

typing behavior, establishing an approach to recog-

nize users by their keyboard usage with an accuracy

of 98% (Alshehri et al., 2016). Besides that, duration

of keystrokes and the time passed between keypresses

were used to get to know speciﬁc people’s typing pat-

terns to create a system using passwords hardened by

personal typing dynamics (Monrose et al., 2001).

Many studies prove that keyboard usage charac-

teristics are personal enough to use them as the base

of authentication when creating an additional layer of

security. When such methods work with these cha-

racteristics, they determine if two patterns match by

calculating how identical they are; that means they

show the matching rate in percentage as a result. Ho-

wever, this value does not show in which parts the ty-

ping characteristics differ in the patterns, and why that

speciﬁc matching rate was calculated. To try to reveal

these differences and similarities in the patterns, we

introduce a visualization method that displays indi-

viduals’ typing styles and shows how speciﬁc typing

346

Enyedi, K. and Kunkli, R.

TypeVis: Visualization of Keystrokes and Typing Patterns based on Time Series Analysis.

DOI: 10.5220/0007584103460353

In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2019), pages 346-353

ISBN: 978-989-758-354-4

characteristics differ among people. By making these

traits visible, we could see in what properties two ty-

ping patterns differ from each other.

2 RELATED WORK

Though many studies use keystroke dynamics as the

source of user authentication, typing characteristics

were presented in a form of visualization as well. In

one study, researchers created a system called Key-

Strokes, which personalizes the typed text based on

such typing characteristics and creates personalized

e-cards (Neumann et al., 2007). Within this system,

handwriting-like personality traits are present in the

case of electronic letters as well. A visualization

method was introduced by the authors by creating a

keyboard-like mapping from some keys, and speciﬁc

key pairs were connected with bubble-like ﬁgures if

they were pressed consecutively. The thickness of the

ﬁgure was determined by the amount of time passed

between the keypresses. Other properties, like the fre-

quency of pressing a key, or the correction of a misty-

ped letter are presented in the background, too. With

all these connected items and represented properties,

the created visualization enriches the typed text that

would be sent on the e-card with personality traits.

There are other kinds of visualization techniques

for characterizing connections among data, too. For

example, using heatmaps is a great idea to present

properties of related items. In a paper, for instance,

a social network visualization system called NetVi-

sia was designed by using heatmaps to display node-

link data (Gove et al., 2011). In another work, the

authors introduced a technique called gesture heat-

maps to aid researchers during gesture analysis (Va-

tavu et al., 2014). Even the already mentioned eye

movement property was visualized using an exten-

sion of the heatmap based technique in another arti-

cle (

Spakov and Miniotas, 2007). In our work, ﬁrst

we tried some visualization methods that are discus-

sed in the next chapter, but our proposed method for

typing visualization also uses heatmap-like visualiza-

tion technology, as discussed in later chapters.

3 OUR FIRST APPROACHES

3.1 KeyStrokes-like Visualization

In our work, we wanted to create a visualization met-

hod that is capable of representing an individual’s ty-

ping style. At ﬁrst, we created an application based on

the ideas of KeyStrokes, trying to extend its features

for our purpose.

Since our goal is to visualize typing styles, in our

application the quantity of the typed text was not limi-

ted, so users could type as much as they wanted. We

noticed that this visualization method is not capable

of showing someone’s typing style, since overlapping

items start to appear on the canvas when the number

of characters starts to grow, making it impossible to

distinguish connections among key pairs (see Figure

1). To overcome this issue, we tried to arrange keys

circularly and also tried to interactively differentiate

connections using colors, but neither of these worka-

rounds helped to create a more comprehensible visu-

alization.

Figure 1: The KeyStrokes-like visualization gets harder to

interpret when the text gets longer. More character means

more connection to draw, which results in a visualization

containing many items that overlap each other.

3.2 Circos-like Visualization

There is a popular genome-based visualization tool

called Circos (Krzywinski et al., 2009) and its web-

based table data visualization tool called Circos Ta-

ble Viewer (Krzywinski, 2008), which works for big-

ger data sets as well. Circos arranges header data of

the table onto a circle and uses ribbons with different

widths to represent the magnitude of the value in the

intersecting cell. If we treat all the keystrokes as a

series of key pairs, then we can group this sequence

by key pairs and represent the data speciﬁc to each

of them. This way visualization items do not over-

lap each other, but it turns out there is another kind

of problem here. Since only those key pairs appear

in the visualization that are present in the typed text,

therefore widths of the ribbons are inﬂuenced by the

number of different key pair combinations. Besides,

if all possible key pair combinations would be shown

at once, then all ribbons would be too thin, and thus

no information could be extracted from the visualiza-

tion (see Figure 2).

TypeVis: Visualization of Keystrokes and Typing Patterns based on Time Series Analysis

347

Figure 2: Circos-like visualization. Connecting ribbons in-

dicate the time passed between each keypress. This method

cannot be used in our context as we need the ability to glo-

bally compare values independently from the number of key

pairs occurring in the text.

3.3 Visualization with B

ezier Curves

Based on our previous visualization attempt using the

idea of Circos, we worked with the same scheme as

we aligned the examined characters in a circle, and

the connection between a character pair held informa-

tion about that keystroke pair. However, this time we

use B

ezier curves as connections instead of ribbons.

More time means higher curvature of the curve at the

point calculated for parameter 0.5 and vice versa. As

can be seen in Figure 3, this results in a visualization

that looks more tense when the user needed less time

to type speciﬁc characters and a bit looser if more.

However, this method involves too much overlap of

the items too, which means it is hard to examine each

element.

(a) Pattern of a faster typer (b) Pattern of a slower typer

Figure 3: Typing pattern of two people visualized with

ezier curves. It can be noticed that items overlap each

other resulting in a crowded diagram.

4 HEATMAP-LIKE

VISUALIZATION

Since in our previous methods, our main problem ba-

sically consisted of the overlapping items, in this kind

of visualization we decided to create space for all pos-

sible key combinations, making a heatmap based re-

presentation. Just like in the case of the Circos-like

method, we treat the typed text as a series of key-

stroke pairs, so we composed a table-like shape that

has all the keys taken into consideration in its row

and column headings, thus the cells are able to re-

present each combination of the keys (see Figure 4).

List of examined characters consists of the letters in

the used language, glyphs, and the space character.

The order of the characters was based on the number

of their occurrences in the selected language, which

means that the alignment of the examined characters

in the visualization is based on their frequency in ge-

neral (Solso and King, 1976). First, we collected the

more frequently appearing characters into the top-left

corner, which means their frequency was decreasing

from left to right and from top to bottom. Then, we

decided to create a rather centralized alignment of the

keys where the more frequently used characters are

placed in the middle of the table. The space charac-

ter is considered as the most frequent one, thus it is

the ﬁrst one occurring in this order, while glyphs are

placed at the end of the list.

By building up the table this way, each cell can in-

form us about the key pair composed of the character

appearing in the row and the column of the cell. For

example, information about the character pair ’ke’ can

be seen in the intersection of row ’k’ and column ’e’.

Figure 4: Central alignment of the examined keys. General

frequency of the characters determine their position in the

table. The more frequent ones appear in the middle, and

when approaching the edges of the table, we encounter less

and less frequent ones. Here, the space is considered as the

most frequent one and glyphs as the least frequent ones.

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

348

4.1 Examined Typing Characteristics

There are many personal typing characteristics that

are speciﬁc to each individual. In this paper, we

only take keystroke dynamics based on time differen-

ces into consideration, which means that no additio-

nal hardware is required to extract these features ex-

cept a regular keyboard that users usually use. In this

section, each typing characteristics that we examine

will be detailed, and we show how they are represen-

ted in the visualization individually. To make it look

clearer, we present them on a smaller table that only

has the ﬁrst 9 characters of the frequency based list.

4.1.1 Flight Time

The most dominant keystroke dynamics we take into

account is the ﬂight time. The ﬂight time is the

amount of time passing between the release of the ﬁrst

and the pressing of the second key in a key pair typed

consecutively, that means this is the time amount be-

tween consecutive keystrokes while no key is being

pressed at all. The amount of ﬂight time can vary in

an interval we deﬁned. By default, we set the bounda-

ries of this interval to 0 and 2000 milliseconds, which

means more than two seconds of ﬂight time will be

considered as a cease in typing and will not be trea-

ted as ﬂight time in the visualization. The value of

the ﬂight time is coded with colors from a color scale

seen in Figure 5.

Figure 5: Possible colors of ﬂight time. Lower boundary of

ﬂight time interval is marked with red and upper boundary

with blue.

It is possible that the interval we set as the default

one for ﬂight time is not the most appropriate one for

an individual’s typing style. One may need more time

than 2 seconds between keystrokes, or it is unneces-

sary to examine time on such a wide interval because

they type a lot faster than that. Therefore, we added

an option to set this interval manually to ﬁt the typing

style of the user, which will be referenced as sensiti-

vity in the followings.

Flight time between consecutive keystrokes is

showed as a square appearing in the corresponding

cells, ﬁlled with a color deﬁned above. Using this

kind of color-coding, higher intensity color shades

mean closer values to the interval boundaries, and as

we can see in Figure 6, red color shades mean shorter

while blue ones mean longer ﬂight time values to in-

dicate the speed and the rhythm of typing with warm

and cold colors. Since users can modify this interval

by setting the above-mentioned sensitivity of the ap-

plication, very similar time amounts can be compared

visually as well.

(a) Shorter ﬂight times (b) Longer ﬂight times

Figure 6: Representing ﬂight times when the user typed the

word ’tea’. Color shades reveal that in (a) ﬂight times were

shorter than in (b), which means the word was typed a little

faster. We also can see in (a) that the character pair ’ea’ was

typed slightly faster than ’te’, since red intensity is a little

higher, and in the case of (b), higher blue intensity of the

character pair ’te’ also indicates that it took more time to

type that than pair ’ea’.

4.1.2 Steadiness of Typing

Steadiness of typing means how much duration values

ﬂuctuate over typing. Since our visualization is based

on averaged values, the extent of how regularly a user

presses and releases speciﬁc key combinations is not

visible. With this in mind, we interpret steadiness of

typing as the actual standard deviation of ﬂight times.

Since this property is based on ﬂight time values, we

illustrate it by setting the size of the colored squares in

each cell. The bigger the standard deviation of ﬂight

times is for a speciﬁc key combination, the smaller

the square will be, which indicates more vacillation

of ﬂight times thus less steady values (see Figure 7).

4.1.3 Negative Flight Time

It can occur that the keyboard user presses a key while

the previously pressed one has not been released yet,

that means there is no ﬂight time between consecu-

tive keystrokes. In this case, there are more keys held

down for a speciﬁc time. We call this duration as ne-

gative ﬂight time and visualize its presence and mag-

nitude in the background of each cell with the shades

of gray color, which means that we actually create

a grayscale frame for each cell and the darkness of

its color determines the amount of the negative ﬂight

time. Frame with a color of white means the lack of

negative ﬂight time, while darker shades mean more

and more time up to 200 milliseconds, which is repre-

sented with black color (see Figure 8).

TypeVis: Visualization of Keystrokes and Typing Patterns based on Time Series Analysis

349

(a) (b)

Figure 7: Representation of typing the word ’tea’ ﬁve ti-

mes separated with spaces. Since these character pairs were

typed multiple times, the average of the ﬂight times is re-

presented in the corresponding cells. It can be seen from

the ﬁgure that some squares are smaller than others, which

means that the user typed those key pairs with higher ﬂuc-

tuation, i.e. higher standard deviation. In (a), ﬂight times

were closer to the average in the case of the ’te’ key pair,

while values were more scattered in the case of typing letter

’a’ and space. In (b), all ﬂight time value are less vacilla-

ting.

(a) (b)

Figure 8: Representing negative ﬂight times appearing

when the user typed once the word ’tea’. It can be seen

that in (a) when typing ’te’ an ’ea’ key pairs, the ﬁrst key of

the pair was not released when the second one has already

been pressed, which means some amount of time passed

while both keys were pressed simultaneously. As we can

see from the darkness of the frames, this time amount was

bigger in the case of ’te’ pair and less for ’ea’. Since the

user only typed once this word and negative ﬂight times ap-

peared, there was no ﬂight time in this case. This also can

be seen as red colors have full intensity. In contrast, in (b),

key pairs have no ﬂight time at all, thus they have no colored

frame.

4.1.4 Duration Time

Similarly to ﬂight time, duration of consecutive key-

strokes can also be measured. Duration time means

the time amount beginning from the moment of pres-

sing the ﬁrst key of the key pair to the moment of

releasing the second key. That means duration time

is the actual ﬂight time appearing at the key pair and

the hold times of the two keys in total. The way of

representing duration time is the same as representing

ﬂight time: colors of the squares show the duration of

the keystroke pair, and the size of the square reveals

the steadiness of duration times occurring by the key

pair. Difference between showing the visualization

with duration and ﬂight times can be seen in Figure

9. This ﬁgure reveals that this property seems to be

a little more stable than ﬂight time is, as values are

more steady and less vacillating. We experienced this

stability effect to be present in typing styles of most

of our users. In the case of showing duration times,

size and color of the cells reveal that values are a little

bigger than ﬂight times are, as they basically mean

more time amount taken into consideration.

(a) (b)

Figure 9: Showing the visualization with ﬂight and duration

times. The user typed the word ’tea’ multiple times. In (a),

ﬂight times and in (b), duration times are presented. The

ﬁrst thing we notice is that ﬂight times are smaller thus they

have warmer colors while duration times last a little longer

as they contain the hold times of the corresponding keys

too.

4.1.5 Typos

It can be questioned if the taken mistakes through ty-

ping could be considered as a personal trait or not. To

examine whether this is an individual trait, we will

display this property on the visualization too. For

now, we assume that when users mistype, they make

a correction using backspace only on the keyboard.

We take into account two things: what was the target

character that the user originally wanted to press, and

what key was accidentally typed instead. We record

the quantity of the target characters and the mistyped

ones, and show them relatively to each other using

bars on the edges of the table. The rate of each tar-

get character can be seen on the bottom of the table

with blue bars, and similarly, the rate for each misty-

ped character is represented with red bars on the right

side of the table, as it can be seen in Figure 10.

We experienced, that it can happen that someone

has a regular typo at a speciﬁc character, but mainly

when people mistype, with a few exceptions they usu-

ally do it with the same dispersion by any character.

Many times they notice the typo after typing a whole

word, which means the space character will be deleted

slightly more most of the time. Besides, this property

depends rather on the text typed by the users, so we

cannot rely on this factor too much. We asked more

people to type the prologue of Shakespeare - Romeo

and Juliet to detect if there is any speciﬁc habit in con-

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

350

nection with mistyping, but we could not notice such

a trait. This is presented in Figure 10.

(a) (b)

Figure 10: Typing patterns of the same user. The user typed

the prologue of Shakespeare - Romeo and Juliet two times,

both are presented in the subﬁgures. Patterns look very si-

milar, so we can guess they are from the same person. The

number of typos, however, is different. We could not notice

any regularity in when and where the user mistyped.

4.2 Unused Cell Elimination

In our heatmap-like visualization method, all possible

key pair combinations examined by us have separated

space, even if no information is present in a cell. In

most of the cases, not all character pairs appear in the

text that the user types, which means many cells re-

main empty when visualizing the typing style of the

user. Since the alignment of the examined characters

is based on their frequency in the used language, va-

lues will not be scattered away too much from each

other but rather appear on a speciﬁc part of the table.

As seen in Figure 11, even after typing a long text,

only about half of the visualization area is ﬁlled with

valuable, non-empty cells. We introduced a method

that eliminates most of the empty spaces. We moved

some of the cells that contain any information into

another position.

If we work with the central alignment of exami-

ned keys, we push each element from all four sides:

from right, up, left, and bottom, alternately. The goal

is to push elements one column and one row closer

to the center in each iteration till we get a row or

column where there is no more empty space, which

means that cells can no longer be pushed one po-

sition further. After applying this process, it is not

clear which key pair belongs to which cell exactly,

but the character frequency based structure still re-

mains because mainly the cells on the edges will be

pushed closer to each other because they appear less

frequently. In contrast, there is more density in the

middle, and those cells cannot be pushed away too

much. This means the process does not violate the

frequency-based structure too much and does not mix

up all the cells. The result of this process can be

seen in Figure 12, where there is a little difference

between the more and the less frequently typed cha-

racters. The process emphasizes non-empty cells by

increasing their size and pushing them closer to ot-

her cells, making them neighboring with cells having

about the same frequency level that they have. Besi-

des, we do not show the mistyping rate for each key

here, since, as discussed above, it is not sure that they

contain properties which are personal enough. After

using this process, a comparison of two different ty-

ping patterns is demonstrated in Figure 13.

5 VALIDATION

For testing the effectiveness of our method, we cre-

ated a survey that involved the help of some volun-

teers. We collected the typing styles of 7 users who

consented to use their patterns for validation purpo-

ses. Everyone had to type the same text consisting

of about 4 500 characters. These patterns were split,

and the ﬁrst and the second halves of them were vi-

sualized separately. In the survey, we presented some

patterns at once that were randomly chosen before,

and we asked the users to pair the patterns that they

think to originate from the same person. All the pe-

ople got the same questionnaire containing the same

patterns. There were 15 questions in the survey. One

type of the questions was about creating two pairs of

four typing patterns presented, and the other kind of

questions were about ﬁnding the matching pair of the

presented pattern. A small portion of this survey is

visible in Figure 14. 16 people participated in ﬁlling

out this form. They were not the same people whose

typing style was presented in the test, and in fact, they

never heard earlier about the existence of the typing

styles of people so both the concept and the presen-

tation method were new for them. Overall, they gave

answers with a correct state rate of 92.08%, showing

that it may be possible to distinguish individuals’ ty-

ping patterns visualized by our method.

6 CONCLUSIONS

We presented a method to visualize individuals’ ty-

ping styles so that personal typing habits can be visi-

ble. In our work, we used time-based keystroke dy-

namics to extract and show the typing patterns of a

person. Since we only rely on time-based properties,

no additional hardware is required except a regular

keyboard when using our software. We experienced

that using time differences of keystroke pairs as the

base of the visualization can result in an individual

TypeVis: Visualization of Keystrokes and Typing Patterns based on Time Series Analysis

351

(a) (b)

Figure 11: Typing style of one person visualized after typing a part of Nathaniel Hawthorne - The Scarlet Letter. The length

of the typed text consisted of about 7 000 characters. In (a), the pattern is presented using ﬂight times and in (b) using duration

times. As we can see from the color intensity difference between the central part and the outer parts of the table, there is a

little difference between typing more and less frequently appearing characters. This pattern is built after typing a long text

that probably has the most of the possible character combinations appearing in the language. However, we can see that about

half of the area has empty cells as those key combinations did not appear in the text typed by the user.

(a) (b)

Figure 12: The same typing style presented before and after

the unused cell elimination process. The user typed a text

in Hungarian, hence Hungarian key mapping is used here.

In (b), with several unused cells removed, we get a more

compact presentation of this typing style.

(a) (b)

Figure 13: Two typing patterns of two different users pre-

sented with ﬂight times, same sensitivity, and unused cell

elimination process. It can be seen in the ﬁgures that main

difference between the two patterns is that in (b), there are

a little longer ﬂight time values than in (a).

and quite distinguishable representation of typing sty-

les of people. Our proposed visualization method is a

heatmap-like technique. We evaluated the results by

creating a validation survey that 16 people ﬁlled out.

They gave correct answers in 92.08% in total.

In the future, some factors should be examined. In

this work, we treated the typed text as a series of key

pairs. However, an examination could be useful to try

using the typing time of three, four, or even more con-

secutive keystrokes, so we might face less vacillation

in time values. Examining how steady the attributes

that we took into account can be useful too, as it is still

unproven whether the typing style of a person chan-

ges over a longer period of time. In addition, we want

to examine if there are any environmental effects that

inﬂuence how a speciﬁc person types. People may

not type the same way when they are tired or nervous

compared to the state when they feel ﬁne. We want to

determine if any of these effects affect the way of ty-

ping, and if they do, how they do it. Checking whether

our proposed visualization method is capable of dis-

playing this kind of factors or not would be relevant.

To ascertain if it is possible for people to detect if so-

meone has such changes by looking at their visualized

typing style would be promising, too.

IVAPP 2019 - 10th International Conference on Information Visualization Theory and Applications

352

(a)

(b)

Figure 14: Two questions taken from the survey. In (a), we

can see a question type where the user had to choose the

corresponding pair. In (b), users had to pair the presented

typing patterns.

ACKNOWLEDGEMENTS

This work was supported by the construction EFOP-

3.6.3-VEKOP-16-2017-00002. The project was sup-

ported by the European Union, co-ﬁnanced by the Eu-

ropean Social Fund.

REFERENCES

Alshehri, A., Coenen, F., and Bollegala, D. (2016). Keybo-

ard Usage Authentication Using Time Series Analy-

sis. In Big Data Analytics and Knowledge Discovery,

pages 239–252. Springer, Cham.

Baker, N. A. and Redfern, M. S. (2007). The Association

between Computer Typing Style and Typing Speeds.

Proceedings of the Human Factors and Ergonomics

Society Annual Meeting, 51(15):869–873.

Bergadano, F., Gunetti, D., and Picardi, C. (2002). User

authentication through keystroke dynamics. ACM T.

Inform. Syst. Se., 5(4):367–397.

Gove, R., Gramsky, N., Kirby, R., Sefer, E., Sopan, A.,

Dunne, C., Shneiderman, B., and Taieb-Maimon, M.

(2011). NetVisia: Heat Map & Matrix Visualization

of Dynamic Social Network Statistics & Content. In

2011 IEEE Third International Conference on Pri-

vacy, Security, Risk and Trust and 2011 IEEE Third

International Conference on Social Computing, pages

19–26. IEEE.

Gunetti, D. and Picardi, C. (2005). Keystroke analysis of

free text. ACM T. Inform. Syst. Se., 8(3):312–347.

Karnan, M., Akila, M., and Krishnaraj, N. (2011). Biome-

tric personal authentication using keystroke dynamics:

A review. Appl. Soft. Comput., 11(2):1565–1573.

Krzywinski, M. (2008). Visualization of tabular data.

http://circos.ca/presentations/articles/

vis_tables1/#vistabular. Access date: Septem-

ber 4, 2017.

Krzywinski, M., Schein, J., Birol,

I., Connors, J., Gas-

coyne, R., Horsman, D., Jones, S. J., and Marra, M. A.

(2009). Circos: an information aesthetic for compara-

tive genomics. Genome Res., 19(9):1639–1645.

Longi, K., Leinonen, J., Nygren, H., Salmi, J., Klami, A.,

and Vihavainen, A. (2015). Identiﬁcation of program-

mers from typing patterns. In Koli Calling ’15 Procee-

dings of the 15th Koli Calling Conference on Compu-

ting Education Research, pages 60–67. Koli, Finland.

Monrose, F., Reiter, M. K., and Wetzel, S. (2001). Password

hardening based on keystroke dynamics. Int. J. Inf.

Secur., 1(2):69–83.

Neumann, P., Tat, A., Zuk, T., and Carpendale, S. (2007).

KeyStrokes: Personalizing Typed Text with Visua-

lization. In EuroVis07: Joint Eurographics, IEEE-

VGTC Symposium on Visualization, pages 43–50.

Norrk

oping, Sweden.

Solso, R. L. and King, J. F. (1976). Frequency and versa-

tility of letters in the English language. Behav. Res.

Meth. Instr., 8(3):283–286.

Vatavu, R.-D., Anthony, L., and Wobbrock, J. O. (2014).

Gesture Heatmaps: Understanding Gesture Perfor-

mance with Colorful Visualizations. In Proceedings of

the 16th International Conference on Multimodal In-

teraction, ICMI ’14, pages 172–179, New York, NY,

USA. ACM.

Spakov, O. and Miniotas, D. (2007). Visualization of Eye

Gaze Data using Heat Maps. Elektron. Elektrotech.,

74(2):55–58.

TypeVis: Visualization of Keystrokes and Typing Patterns based on Time Series Analysis

353