Insights from BB-MAS - A Large Dataset for Typing, Gait and Swipes of the Same Person on Desktop, Tablet and Phone

Insights from BB-MAS - A Large Dataset for Typing, Gait and Swipes of the Same Person on Desktop, Tablet and Phone

Abstract

Behavioral biometrics are key components in the landscape of research in continuous and active user authentication. However, there is a lack of large datasets with multiple activities, such as typing, gait and swipe performed by the same person. Furthermore, large datasets with multiple activities performed on multiple devices by the same person are non-existent. The difficulties of procuring devices, participants, designing protocol, secure storage and on-field hindrances may have contributed to this scarcity. The availability of such a dataset is crucial to forward the research in behavioral biometrics as usage of multiple devices by a person is common nowadays. Through this paper, we share our dataset, the details of its collection, features for each modality and our findings of how keystroke features vary across devices. We have collected data from 117 subjects for typing (both fixed and free text), gait (walking, upstairs and downstairs) and touch on Desktop, Tablet and Phone. The dataset consists a total of about: 3.5 million keystroke events; 57.1 million data-points for accelerometer and gyroscope each; 1.7 million data-points for swipes; and enables future research to explore previously unexplored directions in inter-device and inter-modality biometrics. Our analysis on keystrokes reveals that in most cases, keyhold times are smaller but inter-key latencies are larger, on hand-held devices when compared to desktop. We also present; detailed comparison with related datasets; possible research directions with the dataset; and lessons learnt from the data collection.

Biometric Dataset, Keystroke, Gait, Stairs, Phone, Tablet, Desktop.

1 Introduction

The recent advancements in active or continuous authentication using behavioral biometrics show that they are promising complements to existing one-time authentication methods like PINs, passwords, fingerprint, etc., generally deployed at the entry points of a system. Behavioral biometrics have an implicit usability advantage as the process of authentication is not separate from whatever the user intends to do with the system. In other words, whatever the user activity is, it becomes a mode of authentication. Researchers have explored various modalities such as keystrokes ([disckeyfea, keystrokemon, scankeystroke]), gait ([biogait, gaitwearable, gaitsurvey]), swipes on touch screen ([touchgest, swipegest, toucheval]) to name a few. With growing number of devices used by a person, research in continuous authentication or behavior analysis will have span across devices and activities to stay relevant. However, the scarcity of benchmark datasets for such scenarios are a hindrance. Several attempts have been made to provide benchmark datasets for a single activity like keystrokes ([cmuks, sharedks, laserks, androidks, pressureks, prosodyks, videoks]), gait ([gait1, gait2, gait3, hugadb]) or swipes ([toucheval, touchalytics, touch3, touchgest]) on a single device family like desktop or phone. Few attempts were also made to share benchmark datasets with multiple activities using single device ([unimibshar, har]). However, no benchmark dataset exists for multi-activity in multi-device scenario, where the same activities were performed by the same users on multiple devices. We attempt to fill this gap and provide a benchmark dataset with BB-MAS (Behavioral Biometrics Multi-device and multi-Activity data from Same users) dataset where the same participants have provided typing, gait and swiping data on desktop, phone and tablet.

A total of 117 participants have provided data voluntarily. Each participant has performed typing (including transcription and free text), gait (including walking on a flat corridor, upstairs and downstairs) and swiping using desktop, phone and tablet. The data collection spanned about 3 months and various anonymized demographics information is provided for each participant. The unique ID allocated to the participant is used on all devices and activities.

1.1 Key Contributions:

  • Insights from analysis on how the keystroke feature values vary between desktop, tablet and phone for the same users are presented. We find the keyhold times are smaller and inter-key latencies are larger on hand-held devices when compared to desktop. We posit that, difference in number of contact points while typing (fewer on hand-held device) may lead to such patterns.

  • To the best of our knowledge, a dataset with the typing, gait and touch data from the same users on desktop, tablet and phone is not available publicly at the time of this writing. With data from 117 participants our dataset stands out as unique and rich for exploration in various directions. Each participant’s session ranged between 2 to 2.5 hours, resulting in a total of about: 3.5 million keystroke events; 57.1 million data-points for accelerometer and gyroscope each; 1.7 million data-points for swipes.

  • Features that are commonly described in literature, for data from all activities are described, extracted and shared alongside the raw data.

  • We compare our dataset with other related datasets for keystroke, gait and swipe and highlight their novelty, differences and advantages.

  • Possible research directions using the BB-MAS dataset are discussed and lessons learnt from this elaborate data-collection effort are shared to help future researchers on similar endeavours.

Data collection was carried out between April and June of 2017, after the IRB approval from our university. All participants signed consent forms and have willingly participated in this data collection. All data has been anonymized and any personal identifiers in the data are removed. All subjects, their data and demographic information can only be referenced through the unique participant ID provided to them.

2 Data Collection

The dataset was designed to capture the behavior of the same users performing various day-to-day activities, such as typing, gait and swipes on three commonly used devices such as, desktop, tablet and phone. Activities were deliberately designed to mimic real-life scenarios, for instance, the typing activity consists of both fixed and free text data, the gait activity consists of walking on flat corridors, walking downstairs and upstairs and touch and swipe data consists of activities such as reading and scrolling. The raw data and the features extracted are shared publicly and can be accessed at http://dx.doi.org/10.21227/rpaz-0h66 .

2.1 Devices

Three most commonly used devices type in current times were selected for our data collection. A desktop, tablet and phone would cover most of our modern-day interactions with devices. The details of the devices used in our data collection are as below:

  • Desktops: Two identical desktop stations were setup. Each desktop station consisted of a standard QWERTY keyboard (Dell kb212-b), an optical mouse (Dell ms111-p) and a Dell 21 inch monitor. The keystrokes, mouse movements and clicks were logged.

  • Tablets: HTC-Nexus-9 tablets were used for the tablet section of the data collection. These tablets had a screen size of 8.9 inches, screen resolution of 1536 x 2048 pixels, device dimensions of 9 x 6 x 0.3 inches (Length X Width X Height) and weighed about 435 grams. Keystrokes, accelerometer, gyroscope and touch were logged.

  • Phones: Two different models of phones, Samsung-S6 and HTC-One phones were used in the data collection. The Samsung Galaxy S6 had a screen size of 5.1 inches and screen resolution of 1440 x 2560 pixels with body dimensions of 143.4 x 70.5 x 6.8 mm and weighing 138 grams, whereas the HTC-One had a screen size of 5.0 inches and screen resolution of 1080 x 1920 pixels with body dimensions of 146.4 x 70.6 x 9.4 mm and weighing 160 grams. Keystrokes, accelerometer, gyroscope and touch were logged. The raw data files from different models are identified by the suffix in the file names explained in detail in Section 2.3.

As the default android keyboard does not allow logging of keystrokes, we created and used an android qwerty keyboard on screen which was similar to the default android qwerty keyboard. The phones and tablets were locked in portrait orientation and users were allowed to type on them with any comfortable posture that they preferred. The details of the data collected from these devices and their formats is described in Section 2.3. Figure 1 shows a screenshot of the application with the keyboard for phone. The application on tablet had the same layout but was scaled to match the default keyboard of an android tablet.

Fig. 1: A screenshot from our phone application showing our keyboard to match the default android keyboard.
Fig. 2: The data collection procedure. Tasks a to m were performed by participants in sequence, the corresponding activities and data collected are described in Table I

2.2 How was the data collected?

Emails were sent out to all students, faculty, and staff to procure the participant population. Each participant had to spend two hours on average to perform the set of sequential tasks as illustrated in Fig. 2.

Upon arrival at the data collection location, each participant answered a set of questions that pertained to his/her demographics and technology usage. The participant was then assigned a unique ID and four devices: a desktop, a tablet and two phones (See Table I). The participant then performed the tasks a to m in sequence. a) The participant was asked to sit at the desktop and type two sections of text (fixed-text), ten times each. Each piece of text consisted of two sentences and had an average of 112 characters. The participant was then given a shopping list consisting of six items. They had to use a popular web-browser (Mozilla Firefox) to browse for the best prices for the six items on the list while making notes (on any familiar text editor) about prices, opinions and thoughts. The participant was then given a list of 12 questions of varying cognitive loads (see Appendix A - F) and asked to type their answers in any order he/she preferred for roughly about fifteen minutes. For the entire duration of task a, keystroke and mouse loggers were deployed on the desktop to log all the actions that the participant performed during this task. b) After the completion of task a, the participant was handed a tablet which was running an application where he/she was asked to type the two pieces of static text again followed by a series of ten questions with varying cognitive loads to be answered with a minimum of 50 characters. The questions were placed in a manner that required the participant to swipe vertically and horizontally between questions. For the entire duration of task b, keystroke, touch, accelerometer, and gyroscope loggers were deployed on the tablet to log all typing, swiping, touch, and movement events. After the completion of task b, the participant was asked to place a phone (Phone1) in his/her pants pocket and made to walk in a predefined path while holding the tablet in hand. The path consisted of three doorways and a stairwell, as shown in Figure 2. The tablet displayed buttons to be pressed by the participant before and after passing through a doorway and also before and after taking the staircase. The tasks c, e, and g required the participant to walk, and tasks d and f required the participant to climb downstairs and upstairs respectively. Throughout the tasks c to g, the tablet and the phone (Phone1) logged the accelerometer and gyroscope values. The tablet also logged the pressing of the buttons (doorway and staircase) by the participant.

Upon completion of task g, the tablet was taken from the participant and another phone (Phone2) was handed to them. For task h, Phone2 ran the same application as the tablet in task b, where the participant had to type the two pieces of static text followed by a series of ten questions (not repeated from task b) with varying cognitive loads to be answered with a minimum of 50 characters, requiring the user to swipe between questions. Phone2 logged all keystroke, touch, accelerometer and gyroscope values for typing, swiping, touch, and movement events. Tasks i to m are similar to tasks c to g, differing only in that the participant held Phone2 (instead of the tablet) and Phone1 remained in pocket while performing tasks i to m. Phone1 and Phone2 logged all accelerometer and gyroscope values. Phone2 also logged the pressing of buttons (doorway and staircase) by the participant.

Task Device Activity Data Duration (Approx.)
a Desktop Typing, Browsing Keystroke and Mouse 50 min
b Tablet Typing Keystroke, Swipe, Accelerometer, Gyroscope 25 min
c
Tablet (in hand)
Phone1 (in pocket)
Walking Accelerometer, Gyroscope 5 min
d Climbing down stairs
e Walking
f Climbing upstairs
g Walking
h Phone2 (in hand) Typing Keystroke, Swipe, Accelerometer, Gyroscope 25 min
i
Phone2 (in hand)
Phone1 (in pocket)
Walking
Accelerometer, Gyroscope
5 min
j Climbing down stairs
k Walking
l Climbing upstairs
m Walking
TABLE I: Data collection tasks performed by the participants. For each participant we recorded activities on four devices - a Desktop, a Tablet and two Phones (pocket and hand).

2.3 What is the format of the data?

The raw data from all sensors was originally written to sql databases for speed and accuracy. However, for the convenience of researchers, the raw data and the features extracted from them are organized in simple flat file structure in comma separated format (csv) shared at http://dx.doi.org/10.21227/rpaz-0h66. This section elaborates the organization and format of both raw data files and feature extracted files. Fig. 3 gives an overview of the entire dataset. It is important that the dataset is clearly understood by its researchers for successful research. Therefore, we explain our dataset in great detail in this section.

Fig. 3: Organisation of the files in our dataset.

Raw Data

The raw data from each sensor for each user is stored in folder labelled with the user’s ID. As shown in Fig. 3, folders ”1” to ”117” contain the raw data files for each user, the prefix ID is used to denote the user’s ID. The details and format of the raw data files are as follows:

  • Keystroke Data: The temporal data of every key press and release performed by the subject during tasks a, b and h (Table I) were logged. These files are named;

    • IDDesktopKeyboard.csv

    • IDHandTabletKeyboard.csv

    • IDHandPhoneKeyboard.csv

    accordingly. These files consist four columns, ”EID”: event ID (Integer); ”key”: the key triggering the key-event (String); ”direction”: the type of key-event (Integer, 0 for press and 1 for release); and ”time”: the timestamp of the key-event (String in date-time format with millisecond resolution). Table II provides an example of keystroke files with a snippet from user 1 in our dataset.

    EID key direction time
    0 t 0 2017-04-14 18:09:41.538
    1 t 1 2017-04-14 18:09:41.679
    2 i 0 2017-04-14 18:09:41.819
    .. .. .. ..
    TABLE II: Example for keystroke files from user 1.
  • Mouse Data: In addition to keystrokes, data from mouse usage was also collected during task a (Table I). Please note that there were sampling issues with the mouse data resulting in smaller files, they are included, nonetheless. Mouse events such as, movement, button and wheel were logged into files named;

    • IDMouseMove.csv

    • IDMouseButton.csv

    • IDMouseWheel.csv

    respectively. The Mouse_Move file has six columns, ”EID”: event ID (Integer); ”rX” and ”rY”: the x and y coordinates relative to the active window (Integer); ”pX” and ”pY”: the x and y coordinate on screen (Integer); and ”time”: the timestamp of the mouse-event (String in date-time format with millisecond resolution). The Mouse_Button file has eight columns, six of them are the same as described for Mouse_Move, ”LR”: mouse button (Integer, 0 for left or 1 for right) and ”state”: type of button event (Integer, 0 for press and 1 for release) are the additional columns. The Mouse_Wheel file has seven columns, six of them are the same as described for Mouse_Move in addition to, ”delta”: direction of scroll (Integer, Negative for scroll-down and positive for scroll-up). Tables III, IV and V provide an example mouse movement, button and wheel data respectively, from user 1.

    EID rX rY pX pY time
    0 4 -8 1004 577 2017-04-14 18:09:29.948
    1 8 -14 1919 0 2017-04-14 18:09:30.228
    2 -2 -26 1916 0 2017-04-14 18:21:13.712
    .. .. .. .. .. ..
    TABLE III: Example for mouse movement data from user 1.
    EID rX rY pX pY LR state time
    0 6 -4 1285 242 0 0 2017-04-14 18:21:17.783
    1 -1 3 811 265 0 1 2017-04-14 18:21:21.761
    2 0 0 811 265 0 0 2017-04-14 18:21:22.120
    .. .. .. .. .. .. .. ..
    TABLE IV: Example for mouse button data from user 1.
    EID rX rY pX pY delta time
    0 0 0 1594 708 120 2017-04-14 18:23:10.936
    1 0 0 1545 708 120 2017-04-14 18:23:12.000
    2 0 0 1618 708 120 2017-04-14 18:23:12.575
    .. .. .. .. .. .. ..
    TABLE V: Example for mouse wheel data from user 1.
  • Accelerometer and Gyroscope Data: For tasks from b through m (Table I), the values from accelerometer and gyroscope sensors were logged on suitable devices, such as tablet: for tasks c - g; phone in pocket: for tasks c - g and i - m; and phone in hand: for tasks i - g. The sampling rate for these sensors was about 100Hz. The files with accelerometer and gyroscope from the tablet are named;

    • IDHandTabletAccelerometer.csv

    • IDHandTabletGyroscope.csv

    those from the phone in the pocket are named;

    • IDPocketPhoneAccelerometer.csv

    • IDPocketPhoneGyroscope.csv

    and from the phone in hand are named;

    • IDHandPhoneAccelerometer.csv

    • IDHandPhoneGyroscope.csv

    respectively. The accelerometer files have five columns, ”EID”: event ID (Integer); ”Xvalue”, ”Yvalue”, ”Zvalue”: the acceleration force in m/s on x, y and z axes respectively, excluding the force of gravity (Float); and ”time”: the timestamp of the data point (String in date-time format with millisecond resolution). The gyroscope data have the same five columns, but ”Xvalue”, ”Yvalue” and ”Zvalue” is the rate of rotation in rad/s around x, y and z axis respectively (Float). Tables VI and VII show an example for accelerometer and gyroscope data respectively, from user 1.

    EID Xvalue Yvalue Zvalue time
    0 1.043 3.245 9.087 2017-04-14 18:56:40.215
    1 0.995 3.303 8.936 2017-04-14 18:56:40.216
    2 0.988 3.355 8.880 2017-04-14 18:56:40.234
    .. .. .. .. ..
    TABLE VI: Example for accelerometer data from user 1.
    EID Xvalue Yvalue Zvalue time
    0 -0.045 0.036 -0.013 2017-04-14 18:56:40.440
    1 -0.027 0.027 -0.017 2017-04-14 18:56:40.449
    2 -0.013 0.022 -0.017 2017-04-14 18:56:40.461
    .. .. .. .. ..
    TABLE VII: Example for gyroscope data from user 1.
  • Swipe Data: For tasks b and h, data from swipes were recorded on the tablet and phone in hand respectively. These were logged into files named;

    • IDHandTabletTouchEvent.csv

    • IDHandPhoneTouchEvent.csv

    respectively. The touch data files have ten columns, ”EID”: event ID (Integer); ”Xvalue” and ”Yvalue”: the x and y coordinates on screen (Float), ”pressure”: the approximate pressure applied to the surface by a finger (Float, normalized to a range from 0 (no pressure at all) to 1 (normal pressure)), ”touchMajor” and ”touchMinor”: the length of the major and minor axis, respectively, of an ellipse that represents the touch area (Float, display pixels), ”pointerID”: index of the pointer/touch used in case of multiple touch points (Integer), ”fingerOrientation”: the orientation of the finger in radians relative to the vertical plane of the device (Float, 0 radians indicates that the major axis oriented upwards, is perfectly circular or is of unknown orientation), ”actionType”: indicates the type of event (Integer, 0: finger down/swipe begin, 1: finger up/swipe end and 2:finger move/swipe); and ”time”: the timestamp of the data point (String in date-time format with millisecond resolution).

    ID Xvalue Yvalue pressure touchMajor touchMinor pointerID fingerOrientation actionType time
    0 818.085 1546.25 1.0 0.976 0.488 0 0.0 0.0 2017-4-14 18:56:47:185
    1 819.140 1545.0 1.0 2.929 2.929 0 0.0 null
    2 820.074 1543.893 1.0 2.929 2.929 0 0.0 2.0 2017-4-14 18:56:47:209
    TABLE VIII: Example for swipe data from user 1.
  • Checkpoints Data: For the tasks c - g and i - m, we require checkpoints to separate the data into walking, upstairs and downstairs. The participants were asked to click on buttons on tablet (c - g) or phone in hand (i - m) to mark the opening and closing of doors and start and end of stairs. These checkpoints can be used to separate the data from all other sensors into different activities. Please note that a proctor followed the users during these tasks (making sure not to influence the activity) and noted down incidents where some users clicked the buttons either early or late by a few seconds, adjustments to such timestamps were made manually by adding or subtracting the number of seconds noted down by the proctor. Checkpoint files have three columns, ”EID”: event ID (Integer); ”eventType”: type of event (String, DoorEntry: user at doorway and is about to open door, DoorExit: user has crossed the doorway and the door has closed behind them, StairEntry: user about to start climbing up or down the staircase and StairExit: user has completed climbing up or down a staircase.); and ”time”: the timestamp of the data point (String in date-time format with millisecond resolution). Table IX shows an example of checkpoint data from user 1. Using the checkpoint data, the accelerometer and gyroscope data can be segmented into three; a) between ”DoorExit” and ”DoorEntry” event represents walking on a flat corridor; b) between the first ”StairEntry” and ”StairExit” represents going downstairs; and between second ”StairEntry and ”StairExit” represents going upstairs.

    EID eventType time
    0 DoorEntry 2017-04-14 19:41:45.980
    1 DoorExit 2017-04-14 19:41:50.639
    .. .. ..
    4 StairEntry 2017-04-14 19:42:18.724
    5 StairExit 2017-04-14 19:42:39.105
    .. .. ..
    TABLE IX: Example for checkpoint data from user 1.
  • FreeText Data: In tasks a, b and h, the users had to first transcribe two pieces of fixed text; a) ”this is a test to see if the words that i type are unique to me. there are two sentences in this data sample.”1; and b) ”second session will have different set of lines. carefully selected not to overlap with the first collection phase.”2. The files labelled ”Desktop_FreeText.csv”, ”Tablet_FreeText.csv” and ”Phone_FreeText.csv” provide the timestamp for each user, at which they completed the transcription section and moved to free text section. In our analyses we have considered the entire typing activity as a whole, these files are provided for researchers who may want to separate fixed text and free text for their work.

Features from Raw Data

We extracted popular features that are used in literature for each modality. The feature extraction for our dataset can be grouped into three parts, namely keystroke, gait and swipe features. The files consisting the extracted features have also been included in our dataset. We briefly describe the features and their storage below.

  • Keystroke Features: We select the common twelve unigraphs (single key) and eighteen digraphs (pair of consecutive keys) that occurred the most number of times in all user’s keystroke data. The unigraphs are: ”BACKSPACE”, ”SPACE”, ”a”, ”e”, ”h”, ”i”, ”l”, ”n”, ”r”, ”S” and ”t”. The digraphs are: (’BACKSPACE’, ’BACKSPACE’), (’SPACE’, ’a’), (’SPACE’, ’i’), (’SPACE’, ’s’), (’SPACE’, ’t’), (’e’, ’SPACE’), (’e’, ’n’), (’e’, ’r’), (’e’, ’s’), (’n’, ’SPACE’), (’o’, ’SPACE’), (’o’, ’n’), (’r’, ’e’), (’s’, ’SPACE’), (’s’, ’e’), (’t’, ’SPACE’), (’t’, ’e’) and (’t’, ’h’). For a unigraph we extract the time of the key as a feature:

    • : -

    For a digraph and the following temporal features are extracted:

    • : -

    • : -

    • : -

    • : -

    Fig. 4: Features extracted from keystroke data.

    The figure 4 illustrates the temporal features extracted form keystrokes. These files are stored in folder labelled ”Keystroke_Features” which contains the files names with syntax ”ID_Feature_Name_Device.csv”, where, ID is the user ID (1-117), Feature_Name is the keystroke feature (keyhold, flight1 - flight4) and Device is either desktop, tablet or phone. Each of these files have column denoting the key (”key” in case of keyhold, ”key1” and ”key2” in the case of flight) and a column with the value extracted for the feature.

  • Gait Features: As the raw data for the gait is a pair of signals from the accelerometer and gyroscope we extract features from both. The gait data is further subdivided into three activities; ”Walking” (on a flat corridor); ”Downstairs” (going down the staircase); and ”Upstairs” (going up the staircase). We use a window size of two seconds with a one second overlap between two consecutive windows. For each two second window we extract a host of features from the accelerometer and the gyroscope for x (”Xvalue”), y (”Yvalue”), z (”Zvalue”) and m (m=). A brief description of the features and their column names in files are as follows:

    • Mean: mean of x, y, z and m data denoted , , and respectively.

    • Standard deviation: standard deviation of x, y, z and m data denoted , , and respectively.

    • Band power: band power x, y, z and m data denoted , , and respectively.

    • Energy: energy of the signals x, y, z and m denoted , , and respectively.

    • Median frequency: median frequency of x, y, z and m signals denoted , , and respectively.

    • Inter quartile range: the inter quartile range of x, y, z and m data denoted , , and respectively.

    • Range: range of the x, y, m and z signals denoted , , and respectively.

    • Signal to noise ratio: the signal to noise ratio in x, y, z and m signals denoted , , and respectively.

    • Dynamic time warping distance: the DTW distance between pairs of signals x-y, y-z and x-z denoted as , and respectively.

    • Mutual information: the mutual information between pairs of signals x-y, x-z, x-m, y-z, y-m and z-m denoted as , , , , and respectively.

    • Correlation: the Pearson correlation coefficients between pairs of signals x-y, y-z and x-z signals denoted , and respectively.

    In the ”Gait_Features” folder, we have sub-folders named ”Activity_Features” where activity is either Walking, Downstairs or Upstairs. Each folder consists files with names following the syntax ”ID_Device_In_Hand_ Sensor_Device_Sensor.csv”, where, ID is the user ID (1-117), Device_In_Hand is ”Tablet” for tasks c - g and Phone for tasks i - l, Sensor_Device is the device from which data comes from (HandPhone, HandTablet, PocketPhone) and Sensor is either accelerometer or gyroscope.

  • Swipe Features: For each swipe performed by users on tablet and phone during tasks b and h respectively, various features related to the speed and trajectory of the swipes are extracted. A brief description of the features and their column names in files are as follows:

    • Minimum x and y coordinates: the minimum x and y coordinates in the entire swipe denoted by and respectively.

    • Maximum x and y coordinates: the maximum x and y coordinates in the entire swipe denoted by and respectively.

    • Euclidean distance: the Euclidean distance between the start and end points of the swipe denoted by .

    • Distance list: Euclidean distance between points of a swipe denoted by .

    • Angle: the tangent angle of the swipe denoted by .

    • Time: the total time taken to for the swipe denoted by .

    • Velocity mean and standard deviation: the mean and standard deviation of velocity during the swipe, and respectively.

    • Velocity quartiles: the first, second and third quartiles of velocity during the swipe, , and respectively.

    • Acceleration mean and standard deviation: the mean and standard deviation of acceleration during the swipe, and respectively.

    • Acceleration quartiles: the first, second and third quartiles of acceleration during the swipe, , and respectively.

    • Pressure mean and standard deviation: the mean and standard deviation of pressure during the swipe, and respectively.

    • Pressure quartiles: the first, second and third quartiles of pressure during the swipe, , and respectively.

    • Area mean and standard deviation: the mean and standard deviation of area during the swipe, and respectively.

    • Area quartiles: the first, second and third quartiles of area during the swipe, , and respectively.

    • Direction: the direction of the swipe comparing the displacement of the fingertip in x and y direction, the direction of swipes is deduced as either left, right, up or down denoted by column .

    These files are stored in the directory ”Swipe_Features” and named with syntax ”ID_Device_in_Hand.csv”, where, ID is the user ID (1-117) and Device_in_Hand (Tablet or Phone) is the device on which the swipe was performed.

2.4 Demographics

Each participant was given a unique ID and made to fill out a brief questionnaire consisting questions relating to demographic, physiology and background. This data is stored in the file labelled ”Demographics.csv” in the form of thirteen columns, ”User ID”: unique ID given to each user (Integer); ”Age”: age of the participant in years (Integer); ”Gender”: the gender of the participants (Character, ”F”: Female, ”M”: Male and ”O”: Other); ”Height”: height of the participant in inches (Integer, inches]); ”Ethnicity”: ethnicity of the participant (String); ”Languages Spoken”: languages that the participant can speak fluently (Tuple, [language1, .., languageN]); ”Typing Languages”: languages in which the participant can type (Tuple, [language1, .., languageN]); ”Handedness”: dominant hand for the participant (”Right”, ”Left” or ”Ambidextrous”); ”Desktop Hours”, ”Smartphone Hours” and ”Tablet Hours”: approximate range of hours in a day the participant spends using a desktop, phone and tablet respectively (Range, in hours: 0-1, 2-4, 5-7, 8-12, More than 12); ”Typing Style”: denotes touch and visual typists (Character, ”a”: Do not look at keyboard/Touch typist, ”b”: Must look at keyboard/Visual typist and ”c”: Occasionally look at keyboard/Visual typist); and ”Major/Minor”: participant’s major and minor stream of education (String).

Table X summarizes the demographics of the participants in our dataset. The average age of participants in our study was about 25 years with more than half of the participants aged between 23 to 26 years. The youngest and oldest participants were 19 and 35 years of age respectively. The average height of participants was about 67 inches. The shortest and tallest being 54 and 74 inches respectively. The daily usage hours also reflect the popularity of these devices while desktops and phones appear to be used more than tablets.

Category Size Category Size
Age in years 19 - 22 22
Daily
usage of
desktop
in hours
0 - 1 17
23 - 26 61 2 - 4 58
27 - 30 28 5 - 7 28
>30 06 8 - 12 12
Sex Female 45 >12 2
Male 72
Daily
usage of
phone
in hours
0-1 3
Height in inches
60 6 2-4 51
60-65 40 5-7 43
65-70 43 8-12 16
70 28 >12 4
Spoken
Languages
1 13
Daily
usage of
tablet
0 - 1 93
2 64 2-4 20
3 32 5-7 4
>3 8
Typing
style
Touch 31
Visual 86
TABLE X: Summary of demographic data.

2.5 Where is it stored and how to obtain it?

The entire dataset, feature files and demographic file are hosted at IEEE-Dataport [ourdata]. The url of the dataset is http://dx.doi.org/10.21227/rpaz-0h66 and is open-access complaint, so it can be downloaded with a free IEEE account.

3 Analysis

Through this paper, we aim to provide in-dept description of our dataset. Therefore, we present results from initial research directions that we have explored. We hope the research community will benefit from the dataset and explore the various other directions of research that cannot be addressed in one single research article. We collect the statistics of the keystroke data and compare the average and standard deviation of the keystroke features (Section 2.3.2) across the three different devices. This helps us provide insights about general typing behavior traits on various devices.

3.1 Statistics of keystroke data and insights from feature values across devices

Table XI shows the statistics of keystroke data from our dataset. On an average each participant performed around 11,750, 8,950 and 9,400 keystrokes on desktop, tablet and phone respectively. Even in the minimum condition, each participant has performed 4,350, 4,550 and 5,450 keystrokes on the three devices respectively. When combined, the average keystrokes per user across all three devices is over 30,000 keystrokes and in minimum condition about 19,250.

Desktop Tablet Phone
All
Devices
Average 11760 8952 9395 30153
Stdev 2132 1584 1472 3880
Min 4365 4580 5463 19252
Max 18716 17029 14694 41828
TABLE XI: Keystroke data statistics: Number of keystroke events.

Outlier Detection for Keystroke Features: From the keystroke data, we extract the all temporal keystroke features that are popular in literature (See Section 2.3.2). We use a simple filter to remove any instances of keys that were held down for two seconds or more. We also remove instances of the inter-key pauses that are greater than two seconds. We assume that these were caused by pauses, where the user is either thinking or receiving instructions during the data collection.

Insights from keystroke feature values across devices

To observe how the keystroke features vary across devices, we calculate the average of the average feature values and average of the standard deviation of the feature values from all users in our dataset. To maintain clarity in presentation we use symbols through to denote the digraphs (: (’BACKSPACE’, ’BACKSPACE’), : (’SPACE’, ’a’), : (’SPACE’, ’i’), : (’SPACE’, ’s’), : (’SPACE’, ’t’), : (’e’, ’SPACE’), : (’e’, ’n’), : (’e’, ’r’), : (’e’, ’s’), : (’n’, ’SPACE’), : (’o’, ’SPACE’), : (’o’, ’n’), : (’r’, ’e’), : (’s’, ’SPACE’), : (’s’, ’e’), : (’t’, ’SPACE’), : (’t’, ’e’) and : (’t’, ’h’)). Tables XII to XVI present the values computed for each feature.

Observations: We observe that participants have considerably less keyhold times for all keys on the hand-held devices (tablet and phone). The average of the standard deviation in keyhold time are also very small, less than 50 milliseconds, in all cases except for ”backspace”. However, it is almost completely the opposite when we consider the flight1 to flight4 features from Tables XIII to XVI. We observe that in case of hand-held devices the average feature values are larger than those on desktop. Especially in case of flight1, which is also called the inter-key latency, we can see the values are almost doubled for many digraphs (, , etc.). It is also worth noticing that both the hand-held devices exhibit similar values in most cases and contrasts, if present, are only with the desktop keystroke features.

Insights: From our observations it appears that participants in general, take longer time between keys on phones and tablets when compared to a desktop keyboard. However, once the key is pressed the release event occurs much sooner on the phones and tablets implying that smaller amount on time is spent with the finger on the key. We posit that this occurrence maybe a result of lesser number of fingers being in contact with the typing surface on hand-held devices. In most cases participants type on tablets and phones with just their thumbs compared to their usage of many more fingers for the desktop keyboard thus increasing the keyhold time and reducing the inter-key latency on desktop.

Desktop Tablet Phone
Unigraph (avg) (std) (avg) (std) (avg) (std)
bspace 168 211 128 175 128 129
space 114 57 77 18 89 17
a 137 68 98 22 103 19
e 123 58 85 20 90 18
h 116 53 73 16 81 16
i 119 61 71 16 85 17
l 102 50 71 16 89 19
n 122 63 72 16 83 16
o 118 61 72 15 87 16
r 129 63 78 19 85 17
s 130 60 87 21 94 18
t 116 54 76 20 81 16
TABLE XII: Summary of keyhold feature statistics. All values are in milliseconds.
Desktop Tablet Phone
Digraph (avg) (std) (avg) (std) (avg) (std)
20 258 80 280 20 228
205 247 412 318 360 313
277 247 513 316 472 290
224 245 432 307 433 321
218 240 441 327 408 298
96 167 199 194 162 170
115 151 175 138 144 126
37 110 123 77 130 63
123 114 169 87 162 80
95 137 200 161 186 142
95 110 254 151 236 120
99 110 205 94 181 75
22 92 118 73 121 59
99 163 208 230 163 169
87 97 148 67 147 63
111 158 250 219 198 154
58 94 138 81 136 68
63 88 111 79 121 74
TABLE XIII: Summary of Flight1 feature statistics. All values are in milliseconds.
Desktop Tablet Phone
Digraph (avg) (std) (avg) (std) (avg) (std)
166 169 163 154 173 118
333 248 509 310 462 308
387 246 582 307 556 287
348 247 520 301 529 317
327 238 518 321 488 295
204 165 278 192 253 170
220 156 250 137 231 126
174 114 200 80 217 65
247 119 255 87 253 82
212 148 275 157 274 144
206 117 333 151 326 121
228 122 275 94 263 76
155 93 203 77 209 62
209 165 286 225 255 170
208 103 230 69 238 65
219 158 328 215 288 154
181 101 225 82 224 69
182 97 186 79 202 77
TABLE XIV: Summary of Flight2 feature statistics. All values are in milliseconds.
Desktop Tablet Phone
Digraph (avg) (std) (avg) (std) (avg) (std)
194 156 179 146 187 110
314 245 491 311 450 306
387 248 588 307 559 288
337 245 509 300 524 315
326 236 518 320 496 294
219 167 284 190 252 168
238 154 265 136 236 120
179 112 218 77 230 64
246 122 251 86 248 80
220 145 267 157 266 143
210 114 326 152 320 119
235 121 279 94 268 76
163 93 197 74 208 60
229 162 297 224 258 167
219 100 237 69 246 63
229 157 324 213 277 151
178 93 216 81 220 69
177 90 189 79 203 74
TABLE XV: Summary of Flight3 feature statistics. All values are in milliseconds.
Desktop Tablet Phone
Digraph (avg) (std) (avg) (std) (avg) (std)
380 268 335 230 336 173
441 249 587 302 552 301
495 247 656 299 644 285
459 246 598 294 620 310
436 239 595 313 577 292
327 174 363 188 342 167
343 166 340 135 324 122
315 132 294 82 317 68
369 132 337 89 339 82
336 161 342 155 354 144
320 123 404 150 410 122
362 141 350 95 351 78
296 106 282 79 295 64
339 172 376 222 348 166
340 117 319 72 338 67
337 165 402 210 367 150
300 107 303 83 307 68
296 110 264 79 284 77
TABLE XVI: Summary of Flight4 feature statistics. All values are in milliseconds.

4 Possible Research Directions

  • User authentication for individual devices using keystrokes, gait or swipes: Our dataset provides multiple modalities, activities and scenarios which can be used separately as individual device or activities for user authentication data.

  • Activity recognition: We share data from multiple activities and sub-activities like; free text and fixed text in case of keystrokes; and walking, upstairs and downstairs in case of gait. Recognizing the activity or sub-activity provides better context for methods to be applied.

  • Feature engineering: Many authentication and identification tasks can be improved with a better understanding of feature sets and their effectiveness for each of the modalities.

  • Inter-Device behavior patterns: A unique property of our dataset that sets it apart from openly available datasets is that, the same participants performed many overlapping activities on multiple devices. Therefore, inter-device patterns in behavior in same activity or different activities can be researched. For example, ”Can the typing behavior of a user on desktop reveal their typing behavior on phone?”.

  • Physiological or Demographic information leakage in activities: As we provide a demographic information of each participant, researchers can also explore if the membership of participants in certain demographic group can be identified from different behavioral activities.

  • Demographic menagerie: Existence of groups of users who perform differently at various authentication tasks has been shown in literature [mene1]. Demographic or physiological links to these groupings can be explored.

5 Comparison with other data sets

The datasets that are currently available for behavioral biometrics are collected with the focus on a single activity. We summarize and compare our dataset to other related datasets that are available in literature. We present the key points of comparison with sizeable and related datasets in Tablet XVII.

Dataset
No. of
users
Type of data Type of activity
Device(s)
used by
participants
Highlights

Keystroke

U@B_KD [sharedks] 148 Latencies Fixed & Free text Desktop Four different keyboards used
SBrook_KD [prosodyks] 196 Latencies Free text Desktop Truthful Vs. Deceptive writing
Video_KD [videoks] 30 Latencies & Video Fixed text Desktop Movement and Motor aspects
Pressure_KD [pressureks] 100 Latencies & Pressure Fixed text Desktop Pressure on desktop keyboard
Android_KD [androidks] 42 Latencies & Pressure Fixed text
37 Tablet &
5 Phone users
Addition of pressure features
for Android
Laser_2012 [laserks] 20 Latencies Fixed & Free text Desktop Free vs Fixed text behavior
CMU_KD [cmuks] 51 Latencies Fixed text Desktop
Large number of repititions
by participants
Clarkson_I [clarksonks] 39 Latencies & Video Fixed & Free text Desktop
Video of face and hand
while typing
Clarkson_II[clarkson2] 103 Latencies Fixed and Free text Desktop
Natural and uncontrolled
includes mouse and app data

Gait

Kinematics [gait3] 42
3d Motion Capture
using Force Plates
Overground &
Treadmill
Force Plates
(placed on body)
Anthropometric data &
Pelvis Kinematics
HuGaDB [hugadb] 18
Accelerometer X 6
Gyroscope X 6
EMG X 2
12 Activities
including Gait,
Upstairs and
Downstairs
Accelerometer,
Gyroscope &
EMG sensors
(placed on body)
Body sensor network
data
VideoGait [gait1] 124 Video Walking -
Effect of Viewpoint,
Clothing & Carrying
humanID [gaitchallenge] 122 Video Walking -
12 Experiments with
changes in conditions.
HAR [har] 30
Accelerometer &
Gyroscope
6 Activities
including Gait,
Upstairs and
Downstairs
Smartphone
Activity recognition
using smartphone
UniMiB SHAR [unimibshar] 30 Accelerometer Gait & Fall Smartphone
Activity recognition
& fall detection

Swipe

ASU_Touch[touch3] 75
X, Y coordinates
Pressure
Area of touch
Orientation of finger
Swipe & Touch
Gestures
Smartphone
Re-authentication using
swipe gestures
Touchalytics[touchalytics] 41
Swipe to scroll
through images
Smartphone
Proposed 30
touch features
LTU_Touch[toucheval] 190
Swipe through
questions
Smartphone
Evaluation of verifiers
for touch data
FAST[touchgest] 40 Browsing Smartphone
Fingergestures
Authentication System
using Touchscreen (FAST)

Our Dataset

Keystroke
Same 117 participants
across the datasets
Latencies Fixed and Free text
Desktop, Tablet
& Phone
The same participants
performing multiple, common
day-to-day activities on multiple
devices with real-life placement
and usage of devices
Gait
Accelerometer X 3
Gyroscope X 3
Walking, Upstairs
& Downstairs
Tablet in hand
Phone in hand
Phone in pocket
Swipe
X, Y coordinates
Pressure, Area of touch
Orientation of finger
Swipe through
questions
Tablet
Phone
TABLE XVII: Comparison with other related datasets.

A majority of keystroke datasets are focused on fixed text data with short strings like password, repeated many times by each participant [androidks, pressureks, cmuks]. Most keystroke datasets are collected on desktops [sharedks, prosodyks, cmuks, clarksonks] and very few are on hand-held devices, such as [androidks], which has a only 42 participants of which 37 participants used tablets and only 5 used phones. Such variations limit the usability of datasets. For gait data, there are two datasets that provide sub-activities, similar to our dataset, such as walking, going up and down the stair case [har, unimibshar]. But, both these datasets have only 30 participants. HuGaDB [hugadb], provides a dataset with 12 different activities collected with a body-sensor-network having 6 accelerometers and gyroscopes each and 2 electromyography (EMG) sensors that are placed on the participants body. Though this data approximates the body movements of participants very closely, it would not be suitable for continuous authentication due to the unrealistic placement of senors compared to day-to-day use of phones. In case of VideoGait and humanId ([gait1], [gaitchallenge]), both having over 120 participants, a third party surveillance approach is more suitable as the datasets consists of video recordings of participants gait which is not suitable for on-device continuous authentication.

As touch and swipe as a behavioral biometric is comparatively less explored and the number of datasets is far fewer. All touch and swipe datasets compared in Table XVII collected similar raw data (coordinates, pressure, area and orientation) while using different ways to make participants perform the swipes.

All the datasets discussed above provide data for single activity on single device. How a particular activity from a user varies from device to device, or existence of correlation between different activities on different devices cannot be explored with these datasets. Heterogeneity Human-Activity Recognition dataset [hhar] consists data from multiple devices and multiple movement-related activities (no keystrokes), but as data was collected from only nine users and the device carrying conditions (eight phones carried together in a pouch at the waist and two watches worn on each hand) limit the usability of the dataset for behavioral biometrics.

Therefore, our dataset stands unique by providing data from the same 117 participants performing; a) typing activity, both fixed and free text, on desktop, tablet and phone; b) gait activity, including walking, upstairs and downstairs, with phone in hand, tablet in hand and phone in pocket; and c) swiping activity on tablet and phone.

6 Lessons learnt

The process of collection, curation, pre-processing, storage and sharing of a dataset is indeed challenging. We dealt with various issues at each stage of this effort and share the key lessons learnt, for the benefit of research community, below:

  • Overhead time for the entire process is nontrivial. Time involved in various legalities; preparation and approval of Institutional Review Board (IRB) documents; reaching out and procuring participants; and scheduling them is nontrivial and require great consideration and planning beforehand.

  • Special attention must be given to avoid Personally Identifiable Information (PII) trickling into data. Especially when data collection aims to capture free text, participants might unknowingly divulge personal information such as names, phone numbers, email ids etc., as part of their answers to questions. This can be corrected either in the protocol designing phase (careful consideration to questions) or in the pre-processing phase.

  • An on-site proctor to oversee each participant’s data collection can ensure quality of data especially when data involves capturing key timestamps for activity separation (Section 2.3.2, Checkpoint files). In a few cases we have made manual corrections (by adding or subtracting seconds noted by proctor) to the timestamps where a participant logged them either too early or too late.

  • When data collection involves logging of timestamps on more than one device it is important to make sure the clocks on all devices involved are synchronized to within a few milliseconds of each other. We carried out several test runs to ensure synchronization of the timestamps on all devices, which was challenging as there were four devices (desktop, pocket-phone, hand-phone and tablet) to be used by every participant.

  • Before sharing the data publicly, it is important to represent similar data from different devices in the same format for easier usability. For example; timestamps were logged in a string format (yyyy-mm-dd hr-min-sec.milliseconds) on the desktop and UNIX timestamp on all other devices; and key strokes were logged as characters or keys on the desktop but as ASCII codes on other devices. Therefore, it is better to standardize the data fields before sharing the dataset.

  • Incomplete data can occur from unexpected application or sensor fault or when participants do not complete the entire process. For example, in our dataset, user 117 did not complete the tasks h to m, but other tasks were complete and are included in the final dataset. In rare cases, where there was too little information for a task or activity, it was better to remove the files for completeness of the shared data.

  • In a previous data collection effort for keystrokes [prevdata], we observed some participants tend to fill in low-quality or gibberish text in order to satisfy the minimum text-length criteria (if any) to finish the session earlier. Such occurrences reduce the quality of data and can be remedied either by clearly stating the dos and don’ts to the participants or by the on-site proctor observing and interrupting such behavior.

7 Conclusion and Future Work

Through this paper, we share and provide the details of our large behavioral biometrics dataset for typing, gait and swiping activities of the same user on desktop, tablet and phone (Section 2). The availability of the data on different devices for the same person makes our dataset unique; and with data from 117 participants, also one of the largest. With this dataset researchers can try to explore questions that were not possible with previously available datasets such as; ”Does the typing of an individual on desktop reveal their typing on a tablet or phone? and vice versa” ; ”Can a person’s demographics like age, height, etc., be predicted from the data of typing, gait or swiping activity on any of the devices?”; to name a few. Each of the files in our dataset are described in detail with example snippets for easier visualization and understanding (Section 2.3.1).

We explore, describe, extract and analyze the most popular features for each activity in our dataset. All features are described briefly and also included with the dataset repository (Section 2.3.2). The demographics of the participants is shared and includes various physiological and background information with good spread for most groups (Section 2.4). The analysis of the features reveals interesting insights. We found participants took more time between keys on phones and tablets when compared to a desktop keyboard but the release event was much sooner on the phones and tablets implying that smaller amount on time is spent with the finger on the key (Section 3). As the general style of typing on tablets and phones is with just the two thumbs as opposed to several fingers on desktop, we posit that this occurrence maybe a result of lesser number of fingers being in contact with the typing surface on hand-held devices thus increasing the keyhold time and reducing the inter-key latency on desktop.

This dataset helps address the scarcity in benchmark datasets for multi-device, multi-activity and multi-modality data from the same participants. Collection of a high-quality dataset that can be publicly shared for the benefit of the community, is indeed a tedious and demanding task. Throughout the process, lessons that we have shared in Section 6 are intended for future researchers who make similar endeavors to have an advantage. We also discuss several possible research directions (Section 4) that can be explored with the help of this dataset. As part of our future work we will be exploring these directions.

Appendix A Cognitive Loads [cog1]

Task Level Required activity
Remember 1
Retrieve knowledge from
long-term memory to explain
Understand 2 Explain, summarize or interpret
Apply 3 Apply, execute or implement
Analyze 4
Organize or break material
into constituent parts
Evaluate 5
Critique or make judgments
based on criteria
Create 6
Generate, plan or put elements
together

Appendix B Fixed Text Sentences

  • ”this is a test to see if the words that i type are unique to me. there are two sentences in this data sample.”

  • ”second session will have different set of lines. carefully selected not to overlap with the first collection phase.”

Appendix C Shopping list

  • Mountain Bike

  • Plane tickets from Syracuse, New York to Los Angeles [1 week from today, Coach Seat]

  • Bathing Suit (male or female)

  • Converse All Stair Hiking Boots

  • 24 Pack of Gatorade (24-oz)

  • Ground Transportation (Train, taxi, Bus) from Los Angeles to San Diego [2 weeks from Today]

Appendix D Free text questions on desktop

  • List some of the things that you like about Syracuse University.

  • Which internet browser do you typically use (e.g, Google Chrome, Internet Explorer, Mozilla Firefox, etc.)?

  • What improvements would you like to see in that browser?

  • If you were to draw a picture of Syracuse University, what objects would you include in it?

  • What is your favorite vacation spot? Why do you like to visit there?

  • Give step-by-step driving directions to your favorite restaurant in the Syracuse Area, starting from your dorm room/ home.

  • What hobbies or activities are you involved with outside of school/work? Why?

  • Discuss step-by-step instructions for making your favorite type of sandwich. Write them so that the person who has never done this before can follow your instructions.

  • What television programs do you watch for the news and current events? Why? If you do not watch anything on TV, what is your primary source for news information? What do you like about it?

  • Give a brief, but sufficiently detailed plot description of your favorite book, story, or movie.

  • What social networking websites do you use? What do you like or dislike about these websites? If you do not use any social network, how do you stay in touch with your friends and acquaintances. Why do you not use social networking websites?

  • Who is your favorite actor, actress, singer, comedian, or TV personality? What do you like about them?

Appendix E Free text and multiple-choice questions on phone

  • What type of Smartphone do you typically use?

    • Android

    • iPhone

    • Windows

    • None

    • Other

  • Which best describes you?

    • I have a very active imagination.

    • I take my civic duties, such as voting, seriously.

    • I crave excitement.

    • I would rather cooperate with people than compete with them.

    • I am a worrier.

    • I do not like to talk about myself.

  • Of the courses you’ve taken in college, which was your favorite and why?

  • Think about a class that you did not enjoy. What improvements would you like to see to make the course better?

  • Re-read Question #2 and the responses. Which response do you feel is least applicable to you and why?

  • Do you intend to pursue an advanced degree (e.g., Master’s or Ph.D.)? Why or why not?

  • Find a rule that makes four of the five options alike. Select the option that does not follow this rule:

    • 11.28.45.62

    • 200.217.234.251

    • 192.209.226.243

    • 214.231.248.265

    • 111.127.140.165

  • (Horizontal swipes) Review Question #7 and the answer that you chose. Why was the rule you found/why did you select your answer?

  • What are the topics of Question #6 and Question #10?

  • Give step-by-step directions from this lab space to your dorm room, making specific notes of each time you would descend or ascend stairs.

Appendix F Free text and multiple-choice questions on tablet

  • What type of Tablet do you typically use?

    • Android

    • iPad

    • Windows

    • Amazon Fire

    • None

    • Other

  • Which best describes you?

    • I don’t mind bragging about my skills and accomplishments.

    • I often forget to or neglect to put things back where I found them.

    • I am dominant, forceful, and/or assertive.

    • I am easy-going and lackadaisical.

    • I am set in my ways.

    • I shy away from crowds.

  • What is your ideal job after graduation? Why?

  • Why did you decide to attend Syracuse University?

  • Re-read Question #2 and the responses. Which response do you feel is least applicable to you and why?

  • If all mangoes are golden in color and no golden-colored things are cheap, which of the following is true?

    • A. All mangoes are cheap.

    • B. Golden-colored mangoes are not cheap.

    • C. Either A or B are true.

    • D. Both A and B are true.

    • E. Neither A or B are true.

  • Review Question #6 and the answer that you chose. Why did you select your answer?

  • (Horizontal swipes) If Question #6 was changed to read ”If some mangoes are golden in color and no golden-colored things are cheap”, which answer would be correct and why?

  • What are your thoughts on the current U.S. president? Which policies, if any, would you like to see changed and how?

  • Discuss step-by-step the process for sending an email from your Syracuse email account. Write these instructions such that a person who has never done this before can follow your instructions.

  • Please provide any comments that you have about the survey or the experiment thus far.

References

Footnotes

  1. The transcription sentences were selected based on two criteria: (1) inclusion of many frequently used words in the Oxford English Corpus, and (2) encouraging typing activity on both hands (on both sides on the keyboard). Transcription sentences were typed in lower case.
  2. footnotemark:
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
""
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
   
Add comment
Cancel
Loading ...
402564
This is a comment super asjknd jkasnjk adsnkj
Upvote
Downvote
""
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters
Submit
Cancel

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test
Test description