Download the Full Ebook and Access More Features - ebooknice.

com

(Ebook) Cracking the Personality Code: Predicting

Personality from Smartphone Sensor Data & App
Usage Data by Swaviman Kumar, Ivano Bison, Bruno
Lepri, Fausto Giunchiglia
https://ebooknice.com/product/cracking-the-personality-code-
predicting-personality-from-smartphone-sensor-data-app-
usage-data-56173244

OR CLICK HERE

DOWLOAD EBOOK

Download more ebook instantly today at https://ebooknice.com

 Instant digital products (PDF, ePub, MOBI) ready for you
Download now and discover formats that fit your needs...

Start reading on any device today!

(Ebook) Biota Grow 2C gather 2C cook by Loucas, Jason; Viles, James ISBN
9781459699816, 9781743365571, 9781925268492, 1459699815, 1743365578, 1925268497

https://ebooknice.com/product/biota-grow-2c-gather-2c-cook-6661374

ebooknice.com

(Ebook) Matematik 5000+ Kurs 2c Lärobok by Lena Alfredsson, Hans Heikne, Sanna
Bodemyr ISBN 9789127456600, 9127456609

https://ebooknice.com/product/matematik-5000-kurs-2c-larobok-23848312

ebooknice.com

(Ebook) SAT II Success MATH 1C and 2C 2002 (Peterson's SAT II Success) by Peterson's
ISBN 9780768906677, 0768906679

https://ebooknice.com/product/sat-ii-success-math-1c-and-2c-2002-peterson-s-sat-
ii-success-1722018

ebooknice.com

(Ebook) Master SAT II Math 1c and 2c 4th ed (Arco Master the SAT Subject Test: Math
Levels 1 & 2) by Arco ISBN 9780768923049, 0768923042

https://ebooknice.com/product/master-sat-ii-math-1c-and-2c-4th-ed-arco-master-
the-sat-subject-test-math-levels-1-2-2326094

ebooknice.com
(Ebook) Cambridge IGCSE and O Level History Workbook 2C - Depth Study: the United
States, 1919-41 2nd Edition by Benjamin Harrison ISBN 9781398375147, 9781398375048,
1398375144, 1398375047

https://ebooknice.com/product/cambridge-igcse-and-o-level-history-
workbook-2c-depth-study-the-united-states-1919-41-2nd-edition-53538044

ebooknice.com

(Ebook) Guide to Mobile Data Analytics in Refugee Scenarios: The 'Data for Refugees
Challenge' Study by Albert Ali Salah, Alex Pentland, Bruno Lepri, Emmanuel Letouzé
ISBN 9783030125530, 9783030125547, 303012553X, 3030125548

https://ebooknice.com/product/guide-to-mobile-data-analytics-in-refugee-
scenarios-the-data-for-refugees-challenge-study-10796392

ebooknice.com

(Ebook) The Wiley Handbook of Personality Assessment by Updesh Kumar ISBN

9781119173441, 1119173442

https://ebooknice.com/product/the-wiley-handbook-of-personality-
assessment-5674904

ebooknice.com

(Ebook) Selected Sensor Circuits: From Data Sheet to Simulation by Peter Baumann
ISBN 9783658382117, 3658382112

https://ebooknice.com/product/selected-sensor-circuits-from-data-sheet-to-
simulation-46483066

ebooknice.com

(Ebook) Web Data Mining: Exploring Hyperlinks, Contents, and Usage Data by Bing Liu
ISBN 9783540378815, 3540378812

https://ebooknice.com/product/web-data-mining-exploring-hyperlinks-contents-and-
usage-data-42981538

ebooknice.com
 Department of Sociology and Social Research

Master’s Degree in
Data Science

Final Dissertation

Cracking The Personality Code:

A New Frontier In Personality
Prediction

Supervisors Student

Ivano Bison Swaviman Kumar

Bruno Lepri

Fausto Giunchiglia

Academic year 2022/2023

 Permission To Use

In presenting this thesis as part of the requirements for obtaining a Master’s degree from the University
of Trento, I hereby grant permission for the University’s Libraries to make it available for inspection.
Furthermore, I consent to the potential copying of this thesis, either in its entirety or in part, for
scholarly purposes. Such permission may be granted by the professor or professors who supervised
my thesis work or, in their absence, by the Head of the Department in which my thesis research was
conducted.
I understand and acknowledge that any reproduction or publication of this thesis or its components
for commercial gain is strictly prohibited without my written consent. I also expect that proper
acknowledgment and credit will be given to both me and the University of Trento in any scholarly use
of the material from my thesis.
For inquiries or requests related to the reproduction or use of material from this thesis, whether in
whole or in part, please direct your correspondence to the Head of the Department of Sociology and
Social Research at the following address:

Head of the Department of Sociology and Social Research

via Verdi, 26
I-38122 Trento
 Acknowledgments

I extend my gratitude to Dr. Ivano Bison, my supervisor, for giving me the opportunity to pursue this
work, providing guidance along the way, and fostering an inspiring and supportive environment that
made this journey both meaningful and enjoyable. I am grateful to Dr. Bruno Lepri and Dr. Fausto
Giunchiglia, who agreed to co-supervise this work.
My heartfelt appreciation goes out to my parents for supporting my decision to pursue a master’s degree.
I am deeply grateful to my friends, whose companionship made this journey truly unforgettable.

I am grateful to all the benevolent individuals whose selfless acts of kindness, whether through their
advice, shared wisdom, or a word of encouragement, have made a profound impact on my journey.

Lastly, I would like to express my gratitude to all the professors and colleagues who played a part in
this academic endeavor. Thank you all for your contributions.
Contents

Abstract 3

1 Introduction 5

2 Literature Review 9
2.1 Evolution of Personality Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Personality Traits and the Big Five Model . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Smartphone data and Personality traits . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Rationale for Our Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Background 13
3.1 Supervised Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.1 Ordinary Least Square Regression . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.2 LASSO Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.3 Ridge Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.4 Random Forest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.5 XGBoost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.6 Support Vector Machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1.7 K-Nearest Neighbour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.8 Decision Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Baseline Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.1 Mean Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.2 ZeroR Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Model Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3.1 Cross Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3.2 RMSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.3 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Methods 21
4.1 Study Design and Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Ethics and privacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 The Sensor Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4.1 Big Five Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4.2 App Usage Logs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4.3 Bluetooth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.4.4 Ring Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4.5 Step Counter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.4.6 Touch Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.4.7 Music Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.4.8 Screen Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.4.9 Battery Charge Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4.10 Doze Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 Exploratory Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1
 4.5.1 Univariate Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.5.2 Multivariate Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.6 Prediction Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.6.1 Regression Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.6.2 Classification Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5 Results 61
5.1 Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.1 Statistical Analysis (RMSE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.2 Fit Line Plots Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.3 Residual Plots Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.2 Classification Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2.1 Statistical Analysis (Accuracy) . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

6 Discussion and Future Work 79

6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

Bibliography 83

2
Abstract
Human behavior is inherently intricate, often challenging to explain through traditional mathematical
models. To simplify this complexity, researchers frequently develop intermediate psychological models
that capture specific facets of human behavior. These intermediate models, often derived from per-
sonality assessments, undergo validation using established survey instruments and tend to correlate
with observable behaviors. Typically, these constructs are utilized to predict specific, standardized
aspects of behavior.
The advent of novel sensing systems has ushered in an era of remarkably precise behavior tracking,
raising the intriguing question of whether the reverse process is feasible: Can we deduce psychological
constructs for individuals from their behavioral data? Modern smartphones are equipped with an
array of sensors capable of capturing, filtering, combining, and analyzing data to generate abstract
measures of human behavior. The ability to extract personal profiles or personality types directly
from mobile phone data, without requiring participant interaction, holds potential applications in
marketing, as well as in the initiation of social or health interventions.
In this study, our aim is to model a well-established personality inventory—the Big Five framework
[17]. Activities of students were observed over a 2 months period using parameters readily available
from the smartphone sensors of participants. Correlation analyses were performed and Supervised
machine learning algorithms were implemented along with cross validation to make predictions about
their personality traits using smartphone sensor data. The study illustrated that the root mean
squared error was of a magnitude that allows for actionable predictions regarding an individual’s
personality based on smartphone data.

3
1 Introduction
In today’s age of technology, smartphones have become an integral part of our daily lives. We use them
not just for communication but also for entertainment, education, productivity, navigation and so on.
What many don’t realize however is that these devices are also capable of collecting vast amounts of
data about us, even when we are not actively using them. With more than a dozen sensors housed
within them, smartphones can track our movements, measure our physiological responses, and even
monitor our environment.
This wealth of data presents a unique opportunity to gain insights into human behavior and
personality traits. In particular, it offers the potential to measure and predict personality, a crucial
aspect of human psychology that has long been studied only through self-reported questionnaires.
Traditional ways of measuring personality through questionnaires have limitations [21]. They require
a huge amount of time, resources and effort to administer the tests and there still remains the potential
for bias. Such data collection can be biased due to social desirability bias or the individual’s own lack
of self-awareness. Hence, smartphones provide a new approach for researchers to measure personality
in a more objective and passive manner leveraging vast amounts of sensor generated data [36].
Some of the conventional ways of understanding personality traits are using the Myers-Briggs
Type Indicator (MBTI) [32], Big Five Personality Inventory [17] or NEO personality inventory [9].
The MBTI approach was based on Carl Jung’s theory of personality types and includes 16 different
personality types based on four dichotomies: Extraversion vs Introversion, Sensing vs. Intuition,
Thinking vs. Feeling and Judging vs. Perceiving. This test was used to help individuals understand
their own preferences and how they interacted with others, and could be used in personal development,
team building, and leadership training. The NEO personality inventory measured an individual’s
personality across five dimensions such as openness, conscientiousness, extraversion, agreeableness,
and neuroticism. The NEO-PI consists of 240 questions. Later a more compact approach gained
popularity which was also based on the same five factor model (FFM) and was named The Big
Five Inventory (BFI). It used only 44 questions instead of 240 questions in the NEO-PI approach.
Another trade off between the two approaches is that the NEO-PI measures more specific facets within
each dimension, whereas BFI assesses only the basic level of each dimension. Though the Big Five
framework comes with its own flaws and has been subjected to criticisms on several instances for its
inability to capture overall behavioral characteristics, this framework remains one of the most widely
accepted inventory with consistent results across populations [45].
Based on the Big Five framework, every individual’s personality consists of five latent dimensions,
such as Openness, Conscientiousness, Extraversion, Agreeableness and Neuroticism. The definitions
and descriptions of these Big Five personality traits are as follows:

1. Openness - This trait refers to a person’s openness to new experiences, ideas, and ways of
thinking. Open individuals tend to be imaginative, curious, and creative. People who score high
in Openness may be more willing to challenge traditional ways of doing things and to think
outside the box. People who score low on this trait tend to be more practical and focused on
the present.

2. Conscientiousness - This trait refers to a person’s level of organization, responsibility, and self-
discipline. Conscientious individuals tend to be reliable, hardworking, and detail-oriented. This
is also called the orderliness dimension. People with high Conscientiousness happen to be very
orderly and organized. People who score low on this trait tend to be more laid-back and less
focused on achieving specific goals.

3. Extraversion - This trait refers to a person’s level of sociability, outgoingness, and assertiveness.

5
 They enjoy being around people and tend to be energized by social interactions. They may be
seen as talkative and enthusiastic. People who score low on this trait tend to be more introverted
and prefer quiet, solitary activities.

4. Agreeableness - This trait refers to a person’s level of cooperativeness, empathy, and kindness.
Agreeable individuals tend to be friendly, compassionate, and willing to compromise. People
who score low on this trait tend to be more competitive and may prioritize their own needs over
the needs of others.

5. Neuroticism - People who score high on this trait tend to experience more negative emotions,
such as anxiety, stress, and sadness. They may be more sensitive to criticism and tend to worry
more than others. People who score low on this trait tend to be more emotionally stable and
less prone to experiencing intense negative emotions and more frequent mood swings.

Ideally, an assessment of personality traits should be done in an unobtrusive manner to ensure unbi-
asedness. An assessment is considered unobtrusive if it does not require any attention of the person
being assessed [25, 52]. This not only makes the assessment more convenient for the person being
assessed and can remove subjective bias, but also reduces the risks of measurements being affected by
modified behavior due to the assessed person being consciously aware of the assessment [25]. This is
where modern smartphones come to the rescue, since they come loaded with a host of sensors which
can be employed to unobtrusively gather data about behavior of individuals [11]. Smartphones are a
good option for this kind of study also because they are already widely used and are routinely carried
around by people for most of their day [21]. Physical as well as logical sensors related to location,
communication, phone state (e.g. screen lit, charging status), phone orientation, connections to other
devices and to the internet can be put to use to understand activities like movements, interactions
and daily habits [24].
The purpose of this study is to establish whether individual differences in personality traits can
be detected through data collected from smartphones. For example, users who are actively using
communication applications like whatsapp and telegram may score high on Extraversion. People who
use productivity apps like calculator, calendar and to-do lists may score high on Conscientiousness.
These hypothetical examples serve as guidelines to study the correlation of data with personality.
With the use of a host of exploratory analysis, correlation analysis and machine learning algorithms
plenty of existing research has already linked behavioral indicators derived from smartphone data to
personality traits. However, as far as our knowledge goes, there is no publicly available dataset for
investigating these connections. Therefore, we utilized data from the WeNet study, which was designed
to address this and other research gaps. In our research, we performed an extensive exploratory
analysis, considering self-assessed personality traits and indicators derived from smartphone data.
Using feature selection, we determined indicators that were informative about the personality of
people. We then adopted a predictive approach using linear as well as non-linear models, with a
specific focus on whether combinations of features extracted from various smartphone sensors could
assist in predicting individuals’ personality traits.
My main hypothesis is that an individual’s personality traits, specifically those related to the Big
Five personality traits, can be predicted using real-world behavior data collected from smartphones.

Figure 1.1: Graphical Representation of the Hypothesis

6
 This thesis is composed of 6 chapters. Chapter 2 furnishes the foundational context necessary for
this thesis. It covers previous research related to human behavior using diverse types of technology-
mediated data and also delves into their constraints. It also addresses various types of personality
traits. In Chapter 3, the research background is outlined which includes various machine learning
models, baseline models and model evaluation techniques. In Chapter 4 the experimental configura-
tion for gathering necessary data is outlined, along with thorough explanations of feature extraction
techniques. This chapter also elucidates the methodologies and machine learning models utilized.
Chapter 5 showcases the outcomes, while Chapter 6 offers a summary of the results and the signifi-
cant contributions made in this research.

7
2 Literature Review
2.1 Evolution of Personality Concepts
Personality traits are patterns of thought, emotion, and behavior that are relatively consistent over
time and across situations. They can be described with familiar words such as “reliable”, “sociable”, or
“cheerful”, as well as more specialized terms such as “narcissistic”, “authoritarian”, or “conscientious”.
Psychology has developed an impressive and useful technology for assessing personality traits, but
personality assessment is not limited to psychologists: Everybody does it, every day. We all make
judgments about our own personalities as well as of the personalities of people we meet, and these
judgments are consequential [13].
Several non-human animal species also exhibit individual differences in behavioral patterns, indi-
cating possible existence of personality traits that may even predate humanity [18]. It is likely that
humans have long observed these differences among members of their community, even before they
had a means to effectively record or communicate these ideas through writings. However, the specific
ways in which prehistoric people conceptualized these differences, such as those between an individual
who excelled at cave painting and one who was skilled at ensuring the safety of fellow tribe members
during hunting, are likely lost to history.
The earliest known theory of personality can be traced back to the Greek physician Hippocrates
at about 400 B.C. [30]. He suggested a classification of individual temperaments into four types:
Sanguine (people who are optimistic and hopeful), Melancholic (people who are sad or depressed),
Choleric (people who are irascible, i.e. easily angered), and Phlegmatic (people who are apathetic, i.e.
indifferent and passionless).
This four temperament theory was further developed by Wilhelm Wundt, a German physiologist
who is considered the “father of experimental psychology” [4]. According to Wundt’s model, the four
temperaments represent the extreme ends of a two-dimensional space that is defined by the emotional
vs. unemotional and changeable vs. unchangeable dimensions [30].
During the 1920s, Carl Jung, a Swiss psychiatrist, introduced the terms “extraversion” and “intro-
version” to describe different orientations of personality. Although initially ignored by academic psy-
chologists, Jung’s work has endured despite being based on introspective and interpretive techniques
within the psychoanalytic tradition established by Sigmund Freud. Extraversion and introversion are
still considered important principles in modern models of personality [49].
Personality psychology as an academic field began to take shape in the 1930s, with the estab-
lishment of the first journal, “Journal of Personality,” in 1932. Hans Eysenck, a German-British
psychologist, was an early influential figure who, in the late 1940s, introduced a three-dimensional
model of personality that consisted of “extraversion”, “neuroticism”, and “psychoticism”. Initially
personality psychologists struggled to establish personality as a relevant construct because behavior
can change depending on context. But later they found out that a person’s behavior over a long period
of time and in different situations remained consistent and were related to a person’s personality. The
lexical approach proposed by Goldberg in 1982 involved creating a comprehensive list of adjectives
to describe an individual’s character, which led to the discovery of the largely independent Big Five
factors [16]. These factors are highly stable over time and are predictive of important life outcomes
[29, 35]. As a result, the Big Five factors are now commonly employed both within and outside the
domain of personality psychology.

9
2.2 Personality Traits and the Big Five Model
For this thesis, we are defining personality traits as patterns of behavior, thought, and feeling that
remain consistent across various situations. We are using questionnaires to assess personality traits,
which are based on item response theory [19, 41, 28]. In personality psychology, items usually consist
of statements like “They like to take risks.” and the test-taker responds on a Likert scale from “Very
much like me” to “Not like me at all.” This theory assumes that answers given in a test are informative
about hypothetical latent variables that affect the answers. These latent variables cannot be measured
directly, but can be inferred based on directly observed manifest variables. Factor analysis is used to
discover the main dimensions along which people vary, and five factors consistently emerge in various
populations. These are known as the Big Five and include extraversion, openness to experience,
agreeableness, conscientiousness, and neuroticism [29].

2.3 Smartphone data and Personality traits

There are several ways that researchers have attempted to measure personality traits using smartphone
sensors. One approach is to use self-report questionnaires that are administered via smartphone apps.
Another approach is to use data from smartphone sensors to infer information about behavior and
activity patterns, which can then be used to make inferences about personality traits. Researchers
gathered information on smartphone usage by relying on self-reports from participants. They then
examined the relationships between personality traits and these self-reported patterns of smartphone
usage. The dataset used for this analysis consisted of data from 112 participants who provided infor-
mation about their smartphone usage, which was subsequently used to estimate their personality traits
[5]. In the study, the researchers assessed participants’ personality traits using two questionnaires: the
Coopersmith Self-Esteem Inventory and the NEO-FFI (Neuroticism, Extraversion, Openness, Five-
Factor Inventory) questionnaire. Additionally, participants completed an 8-item questionnaire related
to smartphone usage. This questionnaire aimed to capture information about the amount of time
participants spent on various smartphone activities, including making and receiving calls, sending and
receiving SMS messages, playing games, changing ringtone and wallpaper, and other related activities.
The findings, based on regression analysis, indicated that individuals with extroverted personality
traits tended to spend more time on making and receiving calls and changing wallpapers on their
smartphones. Self-reported mobile gaming behaviors were used to identify the personality traits [38].
The results of the regression analysis showed that individuals who scored low on the Agreeableness
personality trait were more likely to use mobile devices for playing games. Subsequently, a number of
researchers delved into the potential for estimating human behavior through the automated extraction
of smartphone sensor data, phone call information, and app usage data. An overview of various
available smartphone sensors and specific areas of psychological research was presented [21]. App
usage logs from smartphones were used to predict human personality [48]. In these studies, the app
logs were collected and app usage categorized based on the type of usage. Usage was put in the
following groups: communications, tools, productivity, games, media, and finance applications, etc.
For instance, user traits were predicted using a snapshot of installed apps. App logs of 200 participants
were collected, and the applications were grouped based on app purpose. An SVM classifier was
employed to successfully infer an individual’s religion, marital status, whether the user is a parent
of small children, and their mother tongue. Personality traits were identified using individuals’ app
adoption [54]. In this study, app installation logs from a total of 2,043 Android users were analyzed
to identify the Big Five personality traits. This identification was done based on the categories of
apps available on the Google Play Store. The Big Five personality values were categorized into three
classes: low, medium, and high. To model personality traits, a random forest classifier was employed.
The results showed that the model was able to predict personality traits with a 50% success rate. It’s
important to note that this prediction was solely based on app adoption, meaning it considered which
apps were installed by users but did not take into account how those apps were actually used.
Phone call behaviors have been employed as another method to estimate an individual’s Big Five
personality traits. In a specific study involving 39 participants, researchers extracted various call and
SMS-related features from phone logs [34]. These features encompassed factors such as call duration,

10
the timing of calls, and the quantity of text messages sent and received. These extracted features
were then utilized to construct a social communication network. To predict the Big Five personality
traits, a supervised learning approach based on Support Vector Machines (SVM) was employed. The
results of this analysis yielded mean squared errors ranging from 0.73 to 0.86 on a 7-point scale. In
another comprehensive approach, researchers harnessed both standard mobile phone information and
GPS data to predict individuals’ personality traits [31].
They collected conventional carrier logs, including phone calls and text messages, from a group of 69
participants. From this data, they calculated the entropy of calls and texts and also assessed the inter-
event time between text messages and calls. In addition, GPS data was utilized to determine the radius
of gyration, daily travel distances, and the number of distinct places visited by each participant. The
participants’ self-reported Big Five personality traits were categorized into three classes: low, average,
and high. To build predictive models, they employed a Support Vector Machine (SVM) classifier.
Using a ten-fold cross-validation approach, they were able to identify Openness, Conscientiousness,
Extraversion, Agreeableness, and Neuroticism traits with the following accuracies: 49% for Openness,
51% for Conscientiousness, 61% for Extraversion, 51% for Agreeableness, and 63% for Neuroticism.
The researchers also noted correlations between specific personality traits and mobile phone behavior.
For instance, they found that Extraversion and Agreeableness traits were associated with the entropy
of participants’ contacts, suggesting a connection between these personality traits and the diversity
of their social interactions. Furthermore, the variance in the time intervals between phone calls was
correlated with the Conscientiousness trait, indicating a potential link between conscientiousness and
communication patterns.
A study of similar nature, but with a considerably larger sample size, incorporated Bluetooth
sensor data to further explore the prediction of Big Five personality traits [33]. This extensive study
involved the collection of mobile data, including telecommunication data (calls and texts), GPS data,
and Bluetooth sensor logs, from a substantial group of 636 students over a period of 24 months. From
this dataset, various features were extracted. These features encompassed aspects such as face-to-
face contacts or physical proximity to others, as determined by Bluetooth signal strength, as well
as geo-spatial mobility patterns, and the analysis of text messages, calls, and social network friends’
contact lists. The Big Five trait values were categorized into three classes: low, medium, and high.
Researchers utilized Support Vector Machines (SVM) to build models for predicting personality traits.
However, the study reported successful identification only for the Extraversion trait.
Touch screen swipe behaviors have been employed as a means to identify personality traits. In
a study involving 98 participants, researchers collected data on touch screen swipes and extracted
various touch/swipe-related features [1]. These features included parameters such as average velocity,
mean pressure, mean finger area, and others. To predict specific personality traits, the researchers
utilized the self-reported Eysenck personality questionnaire in combination with the extracted touch
screen swipe features. They employed machine learning classifiers, specifically K-Nearest Neighbors
(KNN) and Random Forests, in their analysis. The outcome was the prediction of the Extraversion and
Neuroticism traits with an average accuracy of 62.9%. In a related study, smartphone data, including
call logs, SMS logs, Bluetooth scans, and app usage, was used to predict Big Five Personality traits
in 83 participants over 8 months [7]. Features like Bluetooth IDs, call durations, unique contacts,
SMS length, and app usage were extracted. Personality traits were categorized as low and high,
and a Support Vector Machine binary classifier achieved accuracies of 69.3% for Openness, 74.4%
for Conscientiousness, 75.9% for Extraversion, 69.6% for Agreeableness, and 71.5% for Neuroticism.
The study also found correlations between personality traits and smartphone usage, e.g. Extraversion
correlated with internet usage, while Conscientious individuals used media apps less, and Extroverts
spent more time on calls.
Another study involved 32 participants who provided data from various sources, including appli-
cation usage logs, phone calls, SMS messages, email messages, and self-reported mood states collected
four times a week [27]. An application named Moodscope was developed as a tool for detecting and as-
sessing mood based on smartphone usage, and it successfully demonstrated that mood can be inferred
from sensor data. Researchers employed a multi-linear regression model to analyze this dataset and
determine participants’ mood. Remarkably, the study achieved a successful inference of participants’

11
mood with an accuracy rate of 66%.
A study made use of app usage data, geospatial records (university arrival time and exit time) and
behavioral parameters (such as charging time) collected from 80 students over 1461 days to estimate
the personality inventory of a participants in an unobtrusive manner without the need of parsing
the app-specific content and social media content [24]. This underscores the potential for utilizing
smartphone data to assess and monitor individuals’ emotional states and well-being.

2.4 Rationale for Our Study

There are several limitations to the previous studies.
• Numerous studies have relied on questionnaires and surveys to gather data, offering psychological
insights that may not be directly accessible through observed behavior [12, 20, 22, 42].
• Certain approaches employ pervasive methods that involve analyzing personal email and social
network activities, such as Facebook likes and the number of friends. However, these methods
can potentially violate privacy laws or run counter to research ethics guidelines [2, 34, 15, 40].
• Certain implementations necessitated data such as the number of initiated or received phone
calls, call response rates, phone contacts, SMS usage details (including the number of messages
sent and message response rates), and social networking activities. However, this data is often
proprietary and accessible only to service providers and social networking companies [2, 7, 31,
34, 15, 43].
• Several modern studies have modeled human personality through statistical analysis and classi-
fication models. They achieve this by transforming standard continuous personality trait values
into discrete categories [1, 3, 7, 31, 23, 39, 43, 54].
• Numerous studies, with the exception of [24, 44], which employed machine learning models to
predict human personality, often lack a thorough presentation of model performance. Instead,
they tend to express results solely in terms of statistical metrics, without offering baseline models
for comparison. This absence of baseline models raises questions about the reliability of their
findings [7, 31, 33, 39].
In contrast to these, in this study,
• We utilized an unobtrusive sensing method to collect data, primarily focusing on easily accessible
smartphone data. This included aspects like app usage (excluding internal app data), hardware
sensors like Bluetooth, and software sensors such as screen events, charge events, and step coun-
ters. We specifically collected data on app usage, such as the number of times communication
apps were used. Importantly, this data can be readily obtained by any app developer with the
users’ consent, eliminating the need to depend on social networking companies or network service
providers.
• In contrast to previous studies that relied on a much smaller number of sensors for their analysis,
we employed a total of nine different sensors, including both hard and soft sensors. In conjunction
with the Big Five Inventory (BFI) data collected through questionnaires, this approach enabled
us to create diverse combinations of features extracted from various sensors. This allowed us to
conduct various correlation analyses and predictions using machine learning algorithms
• While the majority of contemporary studies treated the prediction of personality traits as classifi-
cation problems, where continuous personality trait values were transformed into discrete classes
based on value ranges, this study took a distinct approach. In our research, along with classifi-
cation models we also employed regression models, which offered a continuous representation of
personality within the Big Five model, presenting a different perspective on the subject.
• At the end, the results are presented with a comprehensive analysis of model performance, which
includes an appropriate baseline model and an examination of fit behavior through residual
analysis. The use of a baseline model for comparison is uncommon in extant research.

12
3 Background
3.1 Supervised Machine Learning

3.1.1 Ordinary Least Square Regression

Ordinary Least Squares (OLS) is a fundamental statistical method employed in various fields, includ-
ing economics, social sciences, and data science, to explore and quantify the relationships between
variables. OLS is particularly useful when investigating linear relationships, where a dependent vari-
able is assumed to depend on one or more independent variables. The primary objective of OLS is
to estimate the coefficients that define this linear relationship by minimizing the sum of the squared
differences between observed data points and the predictions made by the linear model.
These coefficients reveal the strength and direction of the associations between variables, enabling
researchers to make informed interpretations and predictions. Successful OLS application relies on
meeting certain assumptions, such as linearity, independence of errors, homoscedasticity, normality
of errors, and absence of multicollinearity among independent variables. Therefore, researchers must
carefully assess these assumptions and tailor their OLS models accordingly. OLS not only serves as a
foundational method but also paves the way for more advanced regression techniques, contributing to
robust statistical analysis and informed decision-making across various domains. A linear regression
model with one dependent and one independent variable when plotted along with the regression line
would look like 3.1.

3.1.2 LASSO Regression

LASSO (Least Absolute Shrinkage and Selection Operator) regression is a statistical method used in
linear regression analysis and machine learning for feature selection and regularization. It was intro-
duced by Robert Tibshirani in 1996. LASSO is a variant of linear regression that adds a regularization
term to the traditional least squares regression objective function. The primary purpose of LASSO is
to prevent overfitting and to perform feature selection by shrinking the coefficients of less important
predictor variables to exactly zero.
LASSO adds a regularization term to the linear regression objective function. This term is the
absolute sum of the coefficients multiplied by a tuning parameter (λ):
X X X
min (yi − β0 − βj · xij )2 + λ |βj |

λ (lambda) controls the amount of regularization. A higher λ results in stronger regularization,

which in turn leads to more coefficients being shrunk towards zero.
One of the significant advantages of LASSO is its ability to perform feature selection. As λ
increases, some of the coefficient estimates become exactly zero, effectively excluding the corresponding
predictor variables from the model. This means that LASSO not only provides a predictive model but
also identifies the most important features in the data.
The λ parameter introduces a bias-variance trade-off. A high λ shrinks more coefficients to zero,
which increases bias but reduces variance, while a low λ allows the model to be more complex, reducing
bias but increasing variance. To determine the optimal value of λ, cross-validation techniques are often
used. By testing the model’s performance with different values of λ, you can select the one that gives
the best balance between predictive accuracy and feature selection.

13
 Figure 3.1: Example of a simple OLS Regression.

3.1.3 Ridge Regression

In simple linear regression, you have a dependent variable (Y ) and one or more independent variables
(X). The goal is to find the best-fitting linear equation:

Y = β0 + β1 X1 + β2 X2 + . . . + βn Xn

that minimizes the sum of squared differences between the predicted values and the actual values.
In OLS, the model can become overly complex when you have many features or when there
is multicollinearity (high correlation between independent variables). This complexity can lead to
overfitting, where the model fits the training data very closely but performs poorly on unseen data.
Ridge regression addresses overfitting by adding a regularization term to the OLS objective func-
tion. The objective function of Ridge regression can be defined as:
X
Minimize: RSS + α βi2

Where:

• RSS (Residual Sum of Squares) is the same as in OLS, measuring the sum of squared differences
between predicted and actual values.

• α (alpha) is a hyperparameter that controls the strength of regularization.

•
P 2
βi is the sum of squared regression coefficients. The regularization term penalizes large
coefficients.

The α parameter controls the trade-off between fitting the data well (minimizing RSS) and keeping
the model simple (minimizing the sum of squared coefficients). A larger α leads to a more regularized
model with smaller coefficient values, which is helpful in reducing the impact of multicollinearity and
overfitting.

3.1.4 Random Forest

Random Forests are an ensemble learning method for decision trees. A decision tree is a graph structure
created by splitting the data repeatedly into subsets, usually according to a single feature. A classic

14
decision tree learning algorithm creates a model by splitting along the input dimension of greatest
variance according to a heuristic or a cost function. Essentially, decision trees learn a hierarchy of
(often true/false binary) decisions, leading to a classification of the data. A typical problem with
decision tree learning is its tendency to overfit the data, that is to model the noise in the data as
well as the underlying trend in the data itself, leading to poor model generalizability. To avoid these
overfitting problems, random forests are employed. Random forests, as shown in figure 3.2 divide the
whole data set into random small subsets (without replacement) and the decision tree is constructed
for each subset. An aggregate statistic (usually the mean or mode) of the output of the ensemble
(forest) of decision trees is taken as the actual answer. In regression tasks, Random Forest operates
as a Random Forest Regressor. Instead of predicting discrete categories, it estimates continuous
numerical values. Similar to the classification process, it constructs an ensemble of decision trees but
employs a different aggregation method. Each tree in the ensemble predicts a numerical value, and
the final prediction is obtained by averaging or taking the median of these individual tree predictions.
Random Forest Regression is highly advantageous for capturing non-linear relationships and handling
noisy data while avoiding the pitfalls of over-fitting. Additionally, it provides insights into feature
importance, allowing researchers to identify the most influential variables in predicting the target
variable. This flexibility and robustness make Random Forest a popular choice for both classification
and regression tasks. For an overview of the various decision tree architectures, learning modes and
applications see [26].

Figure 3.2: Sample Random Forests model created with three decision trees for the purpose of
demonstration.

3.1.5 XGBoost
XGBoost, which stands for Extreme Gradient Boosting, is a powerful machine learning algorithm
that falls under the category of ensemble learning. It was developed by Tianqi Chen and is known
for its effectiveness in solving various machine learning problems, especially in structured data and
tabular data scenarios. XGBoost is an implementation of the gradient boosting framework, a machine
learning technique that builds predictive models by combining the predictions of multiple weaker
models, typically decision trees. Gradient boosting works by sequentially training a series of weak
learners and adjusting their predictions to minimize a specified loss function. It’s an ensemble method,
which means it combines multiple models to improve predictive accuracy. XGBoost primarily uses
decision trees as base learners. Decision trees are simple, non-linear models that make predictions
by partitioning the input data into subsets and assigning a constant value to each subset. The
decision trees used in XGBoost are often shallow, with a limited number of nodes, which makes them
weak learners. XGBoost incorporates several techniques to control overfitting and improve model
generalization. Regularization is applied through L1 and L2 regularization terms added to the loss
function, which penalize complex models. This helps prevent the model from fitting the training data

15
 Figure 3.3: Sample XGBoost model created for demonstration

too closely.

While XGBoost uses popular loss functions like mean squared error for regression and log loss for
classification by default, it also allows users to define custom loss functions, making it adaptable to a
wide range of problems. The figure 3.3 shows a demonstration of how XGBoost works.

3.1.6 Support Vector Machine

Support Vector Machines (SVM) represent a powerful and versatile class of supervised machine learn-
ing algorithms used extensively in both classification and regression tasks. In classification, SVM aims
to find a hyperplane that best separates different classes or categories within a dataset. It does so
by maximizing the margin—the distance between the hyperplane and the nearest data points of each
class. SVM is particularly effective in scenarios with complex decision boundaries or high-dimensional
feature spaces, as it can employ various kernel functions (e.g., linear, polynomial, or radial basis
function) to map data into higher-dimensional spaces (as shown in Fig. 3.4 ) where classes become
more separable. This non-linear transformation enables SVM to handle intricate patterns and achieve
high classification accuracy, making it a popular choice in image recognition, text classification, and
bioinformatics.

In regression tasks, SVM transforms into a Support Vector Regressor, aiming to find a hyperplane
that best fits the data while minimizing prediction errors. Unlike traditional regression techniques,
SVM Regression can capture non-linear relationships by utilizing kernel functions to map input fea-
tures into a higher-dimensional space. The objective is to find a hyperplane that maintains a specified
margin around the predicted values, effectively balancing the trade-off between fitting the training
data and generalizing to unseen data points. SVM Regression excels in scenarios where data exhibits
nonlinear patterns, and it is robust to outliers due to its use of support vectors—data points that
influence the position of the hyperplane. As a result, SVM is a valuable tool in both classification and
regression domains, contributing to breakthroughs in fields such as finance, healthcare, and natural
language processing.

16
 Figure 3.4: Visual representation of Support Vector Machine transforming the Non-liner separable
data in to higher dimensional space

3.1.7 K-Nearest Neighbour

The K-Nearest Neighbors (KNN) algorithm is a fundamental and intuitive machine learning technique
used for both classification and regression tasks. It is based on the principle of proximity, assuming
that similar data points in a feature space tend to have similar target values or belong to the same
class. KNN is considered a non-parametric and instance-based learning method because it doesn’t
make assumptions about the data distribution and makes predictions based on local information. To
make a prediction for a new, unseen data point, KNN identifies the k-nearest neighbors from the
training dataset. The distance metric (commonly Euclidean distance or Manhattan distance) is used
to measure the proximity between the new data point and all other data points.

In classification tasks, KNN assigns the class label that is most frequently represented among the
k-nearest neighbors to the new data point. This is often determined by a simple majority vote. In
regression tasks, KNN calculates the average (or weighted average) of the target values of the k-nearest
neighbors and assigns this value as the predicted target value for the new data point.The choice of
the ”k” parameter in KNN is crucial. A small ”k” makes the model sensitive to noise and outliers,
potentially leading to overfitting. A large ”k” can over smooth decision boundaries, potentially leading
to underfitting. KNN is often computationally expensive for large datasets, as it requires calculating
distances between the new data point and all training data points. Various data structures (e.g., KD-
trees) and optimizations can be used to speed up this process.Fig. 3.5 ) shows the Voronoi tessellation
having 19 samples marked with a ”+”, and the Voronoi cell surrounding each sample. A Voronoi cell
encapsulates all neighboring points that are nearest to each sample. For an overview refer to [37].

17
 Figure 3.5: Voronoi tessellation showing a sample k-NN classifier

3.1.8 Decision Tree

Decision tree algorithm is a machine learning technique used for both classification and regression
tasks. It is a non-linear and non-parametric model that builds a tree-like structure to make predictions
by recursively splitting the data based on the most significant features. Each internal node in the
tree represents a feature or attribute, and each branch represents a decision or rule based on that
feature. The leaves of the tree contain the predicted class labels (for classification) or target values
(for regression) for the corresponding subset of data.
The algorithm starts with the entire dataset at the root of the tree. It selects the feature that
provides the best split, typically based on criteria like Gini impurity (for classification) or mean
squared error (for regression). This split partitions the data into subsets, each sent down a branch of
the tree. The process continues recursively for each subset, selecting the best feature for splitting at
each internal node. The splitting stops when a predefined stopping criterion is met, such as reaching
a maximum depth or having a minimum number of samples in a leaf node. Once the tree is built,
the class label (for classification) or target value (for regression) assigned to each leaf node is used
as the prediction for data points that reach that leaf during inference. Decision Trees serve as the
building blocks for more advanced ensemble methods like Random Forests and Gradient Boosting,
which combine multiple decision trees to improve predictive performance and robustness. We have
made use of such ensemble techniques in our work too. Figure 3.6 refers to a sample decision tree
based on binary target variable Y.

Figure 3.6: Sample decision tree based on binary target variable Y

18
3.2 Baseline Models
3.2.1 Mean Model
A Mean model serves as a fundamental baseline model for regression problems. In this simplistic
model, the output value is forecasted solely as the population mean, disregarding any variations in
the input values.
For instance, Table 3.1 includes data for five participants, encompassing three input features and
an output feature known as the ”actual trait.” There is also a column labeled ”predicted trait,” which
represents the values predicted by the mean model. As demonstrated, the mean model consistently
returns a predicted trait value of 0.5 for all participants. This uniform prediction arises because the
mean of the population, in this case, is calculated to be 0.5, and the mean model simply assigns this
value as the prediction for each participant, irrespective of their individual input values.

Participants Input 1 Input 2 Input 3 Actual Trait Predicted Trait

Participant 1 10 5 15 0.8 0.5
Participant 2 7 11 3 0.9 0.5
Participant 3 11 13 17 0.2 0.5
Participant 4 15 8 6 0.5 0.5
Participant 5 3 9 9 0.3 0.5

Table 3.1: Sample Data for Mean Model

3.2.2 ZeroR Model

The ZeroR model, often referred to as the ’zero rules’ or ’most frequent class’ model, serves as a
fundamental baseline model frequently used in classification problems. The ZeroR model adopts a
uniform prediction strategy where the output class label remains constant and equal to the most
prevalent class label observed in the dataset. So in this model, the prediction for the class label is
always the most common class label regardless of the input features or their values.
For instance, consider Table 3.2, which contains data for five participants, including three input
features and output class labels denoted in the ”Actual Trait” column. In this case, the possible
output class labels are ’high’ (H) or ’low’ (L). Since the majority of the population in this dataset has
’H’ as the class label, the ZeroR model consistently predicts ’H’ as the class label for all participants,
irrespective of their individual input feature values. This straightforward approach serves as a baseline
for classification modeling, offering a straightforward way to gauge the performance of more intricate
classification models, offering a clear and easily interpretable comparison in assessing the effectiveness
of advanced techniques and algorithms.

Participants Input 1 Input 2 Input 3 Actual Trait Predicted Trait

Participant 1 10 5 15 H H
Participant 2 7 11 3 H H
Participant 3 11 13 17 H H
Participant 4 15 8 6 L H
Participant 5 3 9 9 L H

Table 3.2: Sample Data for ZeroR Model

3.3 Model Evaluation

3.3.1 Cross Validation
Cross-validation is a technique used to assess the reliability and consistency of machine learning
models. In the process of cross-validation, the entire dataset is divided into K folds, with one fold
reserved as a validation set while the model is trained on the remaining K-1 folds. This approach

19
allows for testing the model against unseen data without introducing potential bias that could arise
from randomly selecting a particularly favorable or unfavorable test set.
For instance, consider a ten-fold cross-validation. The dataset is initially split into ten sets. During
the first round, one set is designated as the validation set, and the model is trained on the remaining
nine sets. This process is then iterated for the remaining sets. The accuracy or error for each round
is computed separately, and the final accuracy or error for the model is determined by averaging the
results from all ten rounds. This comprehensive evaluation provides a robust measure of the model’s
performance across different subsets of the data.

3.3.2 RMSE
The root mean squared error is a quality metric for regression models. It is computed by finding
the square root of the mean of the squares of the difference between the actual values and predicted
values. Since this is an error metric, a model with lower root mean squared error is considered as a
better model. For the sample mean model data shown in Table 3.1 RMSE value is 0.28.
The Root Mean Square Error (RMSE) is calculated as follows:
v
u n
u1 X
RMSE = t (yi − ŷi )2 (3.1)
n
i=1

Where:

n : Number of data points

yi : Actual value for data point i
ŷi : Predicted value for data point i

3.3.3 Accuracy
Accuracy serves as a crucial statistical metric for evaluating classification models. It is calculated by
taking the ratio of the number of correct predictions made by the model to the total number of tests
conducted. A model with a higher accuracy percentage is generally considered to be superior in its
predictive capabilities.
For instance, let’s consider a sample classification dataset as presented in Table 3.2. In this scenario,
the model successfully made three correct predictions out of a total of five tests. Consequently, the
accuracy of the model can be determined as 60% since it achieved a 60% accuracy rate by correctly
classifying the majority of the test cases.

20
4 Methods
4.1 Study Design and Data Collection
The data for this study was collected under a project called WeNet, one of European Union’s Horizon
2020 programs under the Grant Agreement number 823783 [51]. The data collection process spanned
a six-week period and was organized into two stages:

1. The initial synchronic data collection involved the use of three standard close-ended question-
naires. This stage enabled the collection of self-reported general information regarding social
practices.

2. The subsequent diachronic data collection took place through a smartphone app, facilitating the
observation of the daily routines of the students.

As described in Figure. 4.1, the first two weeks were dedicated to the sample recruitment. This
was performed by sending two initial questionnaires, i.e., invitation and assessment of habits. The
remaining month was entirely dedicated to the data collection through the app installed on the students
smartphone. During all the data collection a help-desk was active, and ready to support students in
all the problems which were arising.
The questionnaires were managed with the LimeSurvey platform [47]. Invitations to participate
in the online survey were sent out through LimeSurvey to the email addresses of students enrolled at
various universities. This data collection method was based on the use of Time Diaries, which are a
well-established tool in the social sciences. Time diaries ask respondents to record three key aspects
of their daily lives: the activities they engage in, the locations they visit, and the people they interact
with. Time diaries can be administered in two ways: as ”leave-behind diaries,” where respondents
record their activities in real-time as the day progresses, or as ”recall diaries,” where respondents recall
their activities from the previous day. In our study, we used the iLog app, which allowed students
to provide real-time responses. The questions and answers were structured in accordance with the
HETUS (Harmonised European Time Use Survey) standard [46, 55].
The sample was chosen from the entire student population of the University of Trento. An invita-
tion to participate in the survey was extended to all students, with an initial exclusion criterion applied
to those who did not possess a smartphone compatible with the study (specifically, only Android Op-
erating System versions greater than 5.0) or those who did not regularly attend classes. Subsequent
to the initial contact via email, the online questionnaire was dispatched to inquire about their habits
and routines. The final stage involved the transmission of a password for downloading and installing
the iLog application. This process resulted in 1042 responses. Among these responses, those from
students born after 1993 (with the aim of restricting the involvement of latecomers to the university)
and students who did not actively engage in university life were removed. From the pool of 860 eligible
candidates, a total of 318 students were selected, with the sample size adjusted proportionally to each
department’s student population. This adjustment was made to prevent any misrepresentation of
daily routines stemming from variations in schedules and university sub-communities. Data cleaning
during the data preparation phase resulted in a final dataset that includes information from only 149
participants. This reduction in size, in comparison to the previously mentioned 318 students, is the
outcome of excluding all participants with limited survey participation. Also due to computational
limitations, for this particular work, the data collected between the time period November 2020 and
December 2020 was used which contributed to the reduction in size.

21
 %
Gender
Female 48.7
Male 51.3
Age
<22 47.5
22-26 52.5
Department
Stem 44.57
Non-Stem 55.42
Total 100 (N=149)

Table 4.1: Descriptive statistics of the participants

Figure 4.1: Schematic representation of the study protocol

Table 4.1 shows how the sample is balanced according to the main characteristics, namely gender,
age and departments (whether stem or non-stem) in which the students were enrolled. Furthermore,
it shows the range of annotations given from the participants. The psycho-social characteristics of
the participants are described in Table 4.2. Concerning personality traits (BFI-10), the average of the
scores is between 49.15 and 76.26, with a maximum standard deviation of 23.69 reached in the case of
the Extraversion variable. The range of responses goes from 0 to 100. Students who were enrolled in
the following departments: Engineering and Applied Sciences, Natural Sciences, Medicine and veteri-
nary medicine, and Agricultural, were categorized as belonging to the STEM department. Conversely,
students enrolled in departments such as Social Sciences, Business/Economics, Law, Humanities, and
International Relations and Public Administration were categorized as non-STEM. Apart from psy-
chosocial traits, data was also collected on various other aspects, including (i) daily and extraordinary
journeys, which encompassed the times and means of transportation used; (ii) work routines; and (iii)
study and class attendance routines. In total, 27 questions were posed, resulting in the collection of
78 variables. However, it’s worth noting that for the purposes of this analysis, we did not utilize this
additional data, and therefore, I won’t be elaborating on it further.

4.2 Ethics and privacy

All survey activities and the outcomes achieved at each site adhere to academic and national ethical
standards, prioritizing privacy protection in accordance with applicable laws and guidelines. Further-

22
 mean std median min max
Total Population
Extraversion 49.15 23.69 50.00 0.00 100
Agreeableness 76.26 15.56 75.00 25.00 100
Conscientiousness 64.88 19.27 62.50 12.50 100
Neuroticism 49.70 20.70 50.00 0.00 100
Openness 71.46 18.66 75.00 6.25 100
Female Population
Extraversion 48.72 22.76 50.00 0.00 100
Agreeableness 80.15 14.27 81.25 25.00 100
Conscientiousness 63.69 20.40 62.50 12.50 100
Neuroticism 54.49 19.90 53.13 0.00 100
Openness 70.42 20.37 75.00 6.25 100
Male Population
Extraversion 49.71 24.98 50.00 0.00 100
Agreeableness 71.09 15.77 75.00 31.25 100
Conscientiousness 66.47 17.62 68.75 18.75 100
Neuroticism 43.34 20.10 43.75 0.00 100
Openness 72.84 16.11 75.00 31.25 100

Table 4.2: Descriptive statistics of the psycho-social traits

more, for experiments conducted outside of Europe, the activities and results have been designed to
align with the requirements of a specific European country, as stipulated by the European Commission.
In this context, Italian legislation was chosen as the reference point. Additional details pertaining to
these compliance measures are provided in [14].

4.3 The Sensor Data

The sensor data collected is characterized by its richness and diversity, and no other dataset with
similar properties is currently known to us. Moreover, some of the selected sensors are rather uncon-
ventional. The sensor data can be categorized into two main groups:
1. Hardware (HW) sensors, which encompass sensors such as Bluetooth, Wi-Fi, GPS, and others.
In this study, data from the Bluetooth and Accelerometer sensor were specifically utilized.

2. Software (SW) sensors, referring to all the software events that can be captured from the oper-
ating system and software applications. Examples include events related to Wi-Fi connectivity
and more. A comprehensive list of the software sensors employed in our study is provided in
Table 4.3.
Table 4.3 shows the list of sensors used in the study along with their measurement frequency
and their respective number of observations. Among all the sensors listed in the table, the Step
Counter is a software sensor whose data is derived from data recorded by a hardware sensor called
the Accelerometer. The Step Counter data provides minute-by-minute records of the number of steps
taken by users. Similarly, the Touch Event sensor contains analogous information, documenting the
number of touch events occurring every minute. For the Screen Event, Battery Charge Event, Doze
Event, and Music Event data, each row in the dataset corresponds to a change in the event, recorded
as True/False or On/Off status. The Ring Event data comprises instances when changes in ring
mode occurred, with three registered ring modes: Normal, Silent, and Vibrate. The Bluetooth sensor
captures data regarding the device name, brand, and records of devices in close proximity to the
user’s smartphone. It includes a feature called ”bond” to indicate whether the devices are paired
with the host smartphone or not. This data offers insights into the level of populated areas where
the user is situated at a given time. The data in its existing form could not be used for analysis and,
therefore, needed to be cleaned, transformed, and preprocessed. Furthermore, meaningful features

23
Exploring the Variety of Random
Documents with Different Content
psychographs, etc. Some of the chapter titles are as follows: Practical
methods of substantiating the truths of spiritualism; Testing the
spirits’ sight; Babies, children and adult spirits, reappearing as
children; The gradual development of spirit photography;
Psychographs across ordinary photographs of sitters;
Materialisations. A religious atmosphere pervades the book. The text
is supplemented by fifty-one illustrations, some of them
reproductions of spirit-photographs.

“From a scientific point of view Professor Henslow’s book is utterly

valueless, as it is evident from the opening of his first chapter that he
himself is a spiritualist of the most pronounced type. But as an
extraordinarily definite account of experiments and results with all
the various phenomena of the reputable private seance room, the
book is as marvelous as an Arabian nights’ story and much more
satisfactory because such things actually happened.” C. H. O.

+ − Boston Transcript p6 Mr 31 ’20 580w

“His book, slovenly as it often is in statement, is another moment

in the accumulating mass of evidence which can not be laughed or
sneered or denounced away.”

+ − Review 2:337 Ap 3 ’20 250w

HENSLOW, GEORGE. Religion of the spirit

world; written by the spirits themselves. *$2 Dodd
134

20–15944
 The book is a compilation of famous communications from the
spirit world for the purpose of proving their religious significance.
The author’s object is to show that the life beyond is but a
continuation of life on earth, that we reap what we have sown, that
every character development here on earth counts beyond and that,
in a certain sense, there is a judgment day awaiting us. The contents
are in part: The necessary pre-acquired mental conditions for
securing happiness in the next world; The laws of eternal life; The
gospel of character, preached and practised in the next life; The
acquisition of the Christ-like character and conduct is everything
hereafter, and must be striven for on earth; The true spiritual
meaning of “heaven” and “hell”; The fate of the suicide—a terrible
warning; The nature of man, here and hereafter.

“He gives out matters of opinion constantly as matters of faith. If

such a world as the contributors to this volume depict really existed,
the fact ought to be concealed, in the interests of the preachers of
immortality.” M. F. Egan

− N Y Times p17 S 26 ’20 160w

HENSON, HERBERT HENSLEY, bp. of

Hereford, and others. Church of England; its
nature and its future. *$1.75 Macmillan 283

(Eng ed 20–16630)

“Those who arranged this series of lectures took care to secure a

thoroughly representative group of English clergymen. Their live
lectures taken together set out with considerable force the views of
high, low, and broad churchmen, with two academic
pronouncements from a couple of Oxford professors. The Rev. W. R.
Matthews, dean of King’s college, London, where the lectures were
delivered, in a short preface, states that their purpose was to bring
together exponents of the different tendencies within the church and
to secure from them full and frank statements of their views on the
great problem which gives its title to the book.”—The Times
[London] Lit Sup

Nation 110:773 Je 5 ’20 250w

+ The Times [London] Lit Sup p173 Ap 3
’19 650w

HERBERT, ALAN PATRICK. Bomber gipsy,

and other poems. *$1.50 Knopf 821

(Eng ed 20–1081)

With a few exceptions these poems are reprinted from Punch.

They are spirited and humorous pictures of life at the front. Besides
the title poem some of the pieces are: Ballade of incipient lunacy; The
rest-rumour; At the dump; The atrocity; The ballad of Jones’s
Blighty; The trench code; The mischief-makers; The deserters; Free
meals; The cookers: a song of the transport; A song of plenty.

+ Booklist 17:61 N ’20

 “Because he has a sense of humor, a great deal of common sense
and the good sense to make what is merely good verse and in no way
pretends to be serious poetry, Mr Herbert has given us a very likable
book about the Tommy.” Marguerite Williams

+ N Y Times p24 Ag 22 ’20 100w

+ Springf’d Republican p11a Ag 22 ’20
160w

HERBERT, ALAN PATRICK. Secret battle. *$2

(4c) Knopf

20–628

He was a sensitive, romantic and imaginative lad, lacking

confidence in himself but pathetically eager and conscientious about
doing the right thing, not to make a mess of it, to measure up and
more than measure up to what was required of him. He always
exacted a bit more of himself than could reasonably be expected. He
distinguished himself at Gallipoli in the most trying part of the war
until he was carried down to the ship in a high fever. Later in France,
his record was the same, always doing the over and above his power
of endurance that was bound in the end to undermine his power of
existence. When the strain had become too great and petty jealousies
of fellow officers and the bullying arrogance of the commander had
done their deadly work, the fatal move was made and one of the
bravest men the war knew was shot for cowardice.

“Mr Herbert’s is one of the most interesting and moving English

war books.”
 + Ath p572 Jl 4 ’19 180w

“The story is told with a quiet restraint, with no attempt to pile up

horrors, but with a relentless insistence on the central tragedy. Very
fine work with a limited appeal.”

+ Booklist 16:281 My ’20

Reviewed by H. W. Boynton

+ Bookm 51:78 Mr ’20 580w

“It is simply and vividly told. It reads not like fiction but like fact,
which perhaps it is.”

+ Ind 103:185 Ag 14 ’20 280w

“He evidently and perhaps rightly considered that to draw any

ultimate consequences from his story in the world of conduct would
have diminished its inherent force. That force is very great.”

+ Nation 110:115 Ja 24 ’20 500w

“Very simply, very quietly and naturally, the author builds up the
structure of events, some of them apparently trivial at the time, but
destined later to become of dreadful portent, which at the last
crushes and breaks Harry’s nerve. The logic of it all is unassailable
and perfectly convincing.”

+ N Y Times 25:11 Ja 11 ’20 1100w

 “Vivid, convincing, written in a style at once strong and flexible
and revealing an unusual gift for character portrayal. ‘The secret
battle’ is one of the few really big novels of the world war.”

+ N Y Times 25:190 Ap 18 ’20 200w

“Being the work of a cultivated Englishman, it has the restraint of

the famous public-school tradition. It wishes to betray too little
rather than too much feeling. Its manner is tense with sympathy, but
its matter approaches dryness.” H. W. Boynton

+ − Review 2:257 Mr 13 ’20 350w

Spec 122:800 Je 21 ’19 100w

“The indictment against the verdict is stated quietly and without

passion. The issue it raises is of interest to all ex-service men; how
far must the army treat men as things, how far can and should it
treat them as persons?”

+ Springf’d Republican p6 F 5 ’20 320w

“Needless to say, it is a painful book. Comfortable people who do

not like their feelings harrowed will no more find it to their taste
than they found ‘Justice’ or ‘Jude the obscure’, to their taste. To the
former, indeed, the last part of ‘The secret battle’ offers a striking
parallel. Not in detail, for it is pitched in a quieter key, and its author
expressly states that he is not attempting to indict a system.”

+ − The Times [London] Lit Sup p356 Jl 3

’19 500w
 HERGESHEIMER, JOSEPH. San Cristóbal de
la Habana. *$3 (4c) Knopf 917.29

20–21412

In a passively receptive mood the author went to Havana and

drifted thru his days taking in impressions of the city, of the people,
of the social atmosphere, of its all-pervading romance. “There was
never a more complex spirit than Havana’s, no stranger mingling of
chance and climate and race had ever occurred; but, remarkably, a
unity of effect had been the result, such a singleness as that
possessed by an opera.... It was its special charm to be charged with
sensations rather than facts; a place where facts ... could be safely
ignored.”

“Mr Hergesheimer, translating the spell of Havana into words of

great imagery and color, has visualized its wonderful charm.”

+ Bookm 52:367 Ja ’21 70w

“Half the time we see the city through his meticulously observant
eyes, and the other half he plays Boswell to his own personality and
ideas. The result is an engaging series of vignettes, a most
understanding interpretation, and a remarkably honest human
document.” J. S. N.

+ Freeman 2:478 Ja 26 ’21 230w

“A production at once original and excellent. Mr Hergesheimer

possesses to an extraordinary degree the power of subjectifying the
objective, which is another way of saying that he can make external
realities his very own. In consequence of this happy ability his book
is about one-tenth Havana and nine-tenths Hergesheimer.”

+ N Y Times p22 D 12 ’20 2000w

“Not the least interesting of Mr Hergesheimer’s remarks refer to

the creation of literature, his own and others.’”

+ Springf’d Republican p10 Ja 7 ’20 350w

HERRICK, CHEESMAN ABIAH. Outstanding

days. *$1.25 Am. S. S. union 394

20–4985

A book of selections for readings and recitations for day school and
Sunday school. Each section is prefaced by a discussion of the origin
and meaning of the special day under consideration. “A collection of
nearly a hundred literary selections is presented in connection with
the several studies. Some of these are old favorites which can never
be out of date. Others are relatively recent, furnishing an expression
of the thought and feeling of the present on the subjects discussed.”
Contents: Place of special days; New Year’s day; Lincoln’s birthday;
Washington’s birthday; Good Friday; Easter Sunday; Mother’s day;
Memorial day; Children’s day; Flag day; Commencement day;
Independence day; Labor day; Beginning school; Thanksgiving day;
Christmas day.
 + El School J 20:795 Je ’20 100w

HERRICK, GLENN WASHINGTON. Insects of

economic importance. *$2 Macmillan 632.7

20–12386

These “outlines of lectures in economic entomology” are a revised

edition of a previous volume. Space considerations prevent the
inclusion of all insects of economic importance. “However, the
principal pests of our important fruits, vegetables, cereals, farm
animals, shade trees, and of the household are discussed. A brief
summary of the life habits of each, so far as they are known, is made,
and the latest methods of control are outlined. In addition, a concise
discussion of insecticides is given together with formulæ and
directions for making and applying them.” (Preface) The first twelve
chapters are: Losses caused by insects; Useful insects; Entomological
literature; Natural methods of insect control; Artificial methods of
insect control; Poison insecticides; Poison baits; Contact insecticides;
Fumigating substances; Miscellaneous means of insect control;
Dusting; Quarantine and insecticide laws. The remainder of the book
is devoted to the special insect pests and their victims and an index.

Booklist 17:165 Ja ’21

R of Rs 62:336 S ’20 40w
 HERRINGHAM, SIR WILMOT PARKER.
Physician in France. (Liverpool diocesan board of
divinity publications) *$5 (*15s) Longmans 940.475

(Eng ed 19–19873)

“Preliminary to this narrative the author discusses the surprise of

the English at the sudden outbreak of the war. After this preliminary
discussion he, in his fifth chapter, begins his personal narrative and
relates the early operations of the medical corps in England at the
beginning of the war, showing us how the thing was done and the
sanitary precautions that were made against sickness among the
forces. Continuing, he tells of the organization and work of the Field
ambulance corps; of the clearing stations; of the work of transporting
the wounded and of the base hospitals and nurses. He then discusses
some phases of medical work, especially the management of cases of
enteric and other fevers, and of shell shock. He talks of the advance
of medicine in the war, of the operations on the plains of Flanders: of
the medical headquarters at Hesdin. Diverging, the author, drawing
from his experiences abroad, tells of education and the religious
question in France and of some interesting contrasts between French
and English people, in domestic manners and management and in
human characteristics.”—Boston Transcript

“The reasons for his popularity will be apparent to anyone who

reads his book, for it exhibits in an attractive form the qualities of his
mind and general outlook.”

+ Ath p1401 D 26 ’19 520w

 + Boston Transcript p10 F 21 ’20 480w

“It is written in ordinary, straightforward language, free from

those amateur attempts at the literary manner which make most
books written by doctors so tedious. Much of the book is political,
and this, except as throwing light on the character of the author, is
the least important part. The most entertaining part of the book
consists in the record of the author’s observations of French life and
its contrasts with our own.” H. R.

+ Nation [London] 26:360 D 6 ’19 1350w

“Entertaining and instructive. The purely medical chapters of the

book have their value as a lucid exposition calculated to enlighten the
layman and to enlist his sympathy.”

+ Sat R 129:37 Ja 10 ’20 1350w

“In the opening chapters, devoted to a consideration of the causes

which led up to the outbreak of the great war, the author exhibits a
fine patriotism tempered by broad-mindedness. The book will
enhance the author’s reputation, and prove most welcome reading
after the publication of so many self-centred memoirs.”

+ Spec 124:245 F 21 ’20 1050w

“Unfortunately, the opening chapters are platitudinous and have

nothing to do with the author’s real theme; but the book improves as
he gets into his stride, and is best of all in the later chapters, devoted
to the differences between the customs and viewpoints of the French
and ourselves, which are handled at once frankly and with
comprehension and discretion.”

+ − The Times [London] Lit Sup p643 N 13

’19 1250w

HEWLETT, MAURICE HENRY. Light heart.

*$2 (5c) Holt

20–8858

This tale is a story of men’s friendships. Thormod, of the light

heart, is a poet who easily wins the love of women, but his real
devotion is given to men, first to his friend Thorgar, whose death he
avenges, then to King Olaf. In his preface the author says, “Of this
heroic, naked story, three fragments survive in ‘Origines Islandicæ,’
that learned repository; but to compound one plain tale of them it
has been necessary to go for the catastrophe to the Saga of King Olaf.
As a result of my hunting and piecing I am able to give an orderly
account of the life of a young man which, I think, justifies the title I
have given it.”

Ath p559 Ap 23 ’20 40w

+ − Booklist 17:33 O ’20

“While ‘The light heart’ is far less interesting and far less stirring
than either ‘Gudrid the fair’ or ‘The outlaw,’ it has one truly splendid
moment—that in which Thormod swears his allegiance for life and
death to King Olaf.”

+ − N Y Times 25:291 Je 6 ’20 900w

“I confess that for me the starkness, the frugality, the astringency

of this tale render it a tougher morsel than some of the Norse fables
Mr Hewlett has previously wrought from similar materials. For his
sources he shows a reverence almost excessive.” H. W. Boynton

+ − Review 3:110 Ag 4 ’20 340w

“The story is good and unusual. But above all we would commend
Mr Hewlett’s short introduction on the nature of the Sagas.”

+ Sat R 130:40 Jl 10 ’20 100w

“The story has retained the legendary atmosphere of the twelfth

century Iceland and Norway. The book is written with Hewlett’s
usual romantic touch. It is interesting mainly on account of the
unusual setting and the strangeness of the characters treated. The
author sacrifices plot to faithfulness to his sources.”

+ Springf’d Republican p11a Je 20 ’20

480w

“Colloquial and prosaic though the telling is—prosaic even in

describing dreams and visions—there shines through it a spirit which
is high and beautiful.”
 + The Times [London] Lit Sup p255 Ap
22 ’20 1000w

HEWLETT, MAURICE HENRY. Mainwaring.

*$2 (4c) Dodd

20–19506

The story portrays two extremely opposed types, a man and a

woman. Mainwaring is a genius of a sort, grasping everything to
himself, ambitious, a demagogue, reckless and unmoral. From
obscurity he rises to political power and is only stayed from
achieving the highest rung by disease and death. He burns himself
out prematurely. While still quite young and out of his mastering
passion of grasping everything he wants, he forces a beautiful young
working girl to marry him. Lizzy in her selflessness, her poise and
sincerity, her obedience to duty, is his opposite. She endures
starvation with him but when he asks her to follow him into high life
she refuses. She has seen through it at a glance and hates it, and
prefers the duties of a housemaid to those of hostess at his banquets.
He subjects her to every indignity but willingly accepts her services
as a nurse during his last days.

Booklist 17:116 D ’20

“Mainwaring stands before a dull gray background, which is rather

bad for the story, but serves the purpose of the novelist in making
Mainwaring a crimson figure against this same gray. As usual, Mr
Hewlett is fascinatingly facile with his pen, but this same smooth
style cannot wholly atone for a very flimsy plot and a succession of
avowed characters that are of no more use than a Greek chorus.”

+ − Boston Transcript p7 N 24 ’20 390w

“Lizzy is a human being, strongly conventional in her sense of

duty, yet as freshly natural in emotional values as Eve strayed from
the garden. On the whole, however, ‘Mainwaring’ is a
disappointment as a novel. The author too apparently is doing over
again with unconvincing dexterity things once well accomplished in
‘Rest Harrow’.”

− + N Y Evening Post p22 O 23 ’20 300w

“The sharp contrasts between these well-drawn figures, whose

souls are silhouetted by the tragic circumstances in which the author
places them, afforded Mr Hewlett equal opportunity to display his
powers of creating and analyzing character. The artistry and dignity
of the story he has written around them make ‘Mainwaring’ a worthy
addition to the novels bearing his name.”

+ N Y Times p22 S 26 ’20 560w

“The political part of the story is not excessively interesting,

although it has capital pen sketches of Disraeli and Gladstone under
slight disguises. Like all Mr Hewlett’s writing, the literary execution
of the book is admirable in its finish and quiet effectiveness.”

+ − Outlook 126:333 O 20 ’20 170w

 “A brilliant study in its kind; but some of us will feel as we have
often felt with Mr Hewlett, that the childlike creature woman rather
than the childish creature man gives the story its charm.
Mainwaring’s Lizzy is a girl to be remembered.” H. W. Boynton

+ − Review 3:382 O 27 ’20 340w

“The two characters are analyzed in vigorous fashion and will

stand as examples of Mr Hewlett’s most finished work.”

+ Springf’d Republican p9a O 31 ’20 450w

HEWLETT, MAURICE HENRY. Outlaw.

*$1.75 (3½c) Dodd

20–4

This is the fifth of Maurice Hewlett’s saga tales retold. It is the

story of Gisli and of Grayflanks, the sword on which a curse was laid
when it was turned against its owner. Young Gisli is a craftsman and
man of peace, who nevertheless is fated to be the slayer of men, to
flee from Norway to Iceland, to become an outlaw, and to die fighting
with his back against the wall, his wife, Aud, beside him.

“We cannot help wishing that he had been a great deal more
lenient with himself. For the tale, as it stands, is so exceedingly plain,
and the fights, murders, escapes and pursuits described upon so even
a breath, that it is hard to believe the great, more than life-size dolls
minded whether they were hit over the head or not. There is no
doubt that the very large number of words of one syllable help to
keep the tone low. They have a curious effect upon the reader. He
finds himself, as it were, reading aloud, spelling out the tale.” K. M.

+ − Ath p15 Ja 2 ’20 600w

Booklist 16:244 Ap ’20

“None of his stories out of the Icelandic sagas is as spirited as ‘The

outlaw.’ The vein of romance discovered in them by Mr Hewlett
seems to be inexhaustible.” E. F. E

+ Boston Transcript p6 Mr 24 ’20 1150w

“‘The outlaw’ is a noble tale fully and in the main nobly told.”
Ludwig Lewisohn

+ Nation 111:191 Ag 14 ’20 500w

“A grim tale, full of strong passions and desperate fighting, is this

of ‘The outlaw.’”

+ N Y Times 25:1 Mr 7 ’20 1000w

+ N Y Times 25:190 Ap 18 ’20 70w

“Needless to say, it is masterly in its art and vividness; yet many of

the author’s admirers would welcome his return to that type of
writing that gave us ‘Half-way house’ and ‘Richard yea-and-nay.’”
 + − Outlook 124:563 Mr 31 ’20 60w

“Mr Hewlett tells a tense dramatic story, reveals studious research

of ancient lore and a singular gift for vitalizing the remote scenes of a
vanished civilization. This is no mere approximation of what the
Vikings were and what they did. It is a lifelike recreation.”

+ Springf’d Republican p8a Ap 4 ’20 550w

“In reproducing the old story Mr Hewlett mediates with his usual
skill between the Scylla of excessive modernity and the Charybdis of
an obsolete idiom. It is, however, questionable whether he might not
without harm have ventured even closer to Scylla.”

+ − The Times [London] Lit Sup p649 N 13

’19 600w

HEYDRICK, BENJAMIN ALEXANDER, ed.

Americans all; stories of American life of today.
*$1.50 (1½c) Harcourt

20–14759

The editor of this volume of short stories states in his preface that
he believes that the short story is the form which can best stand as
the adequate expression in fiction of American life. He says “If it
were possible to bring together in a single volume a group of these,
each one reflecting faithfully one facet of our many-sided life, would
not such a book be a truer picture of America than any single novel
could present? The present volume is an attempt to do this.”
Contents: The right Promethean fire, by George Madden Martin; The
land of heart’s desire, by Myra Kelly; The tenor, by H. C. Bunner; The
passing of Priscilla Winthrop, by William Allen White; The gift of the
Magi, by O. Henry; The gold brick, by Brand Whitlock; His mother’s
son, by Edna Ferber; Bitter-sweet, by Fannie Hurst; The riverman,
by Stewart Edward White; Flint and fire, by Dorothy Canfield; The
ordeal at Mt Hope, by Paul Laurence Dunbar; Israel Drake, by
Katherine Mayo; The struggles and triumph of Isidro de los
Maestros, by James M. Hopper; The citizen, by James F. Dwyer.
There is a sketch of the author following each story, and at the end a
List of American short stories classified by locality, and Notes and
questions for study.

“An interesting group of stories.”

+ Booklist 17:158 Ja ’21

Boston Transcript p4 O 9 ’20 280w

“Only two stories in the volume, Myra Kelly’s ‘Just kids’ and
William Allen White’s ‘Society in our town,’ have grown instead of
being made after a model.”

− + Nation 111:692 D 15 ’20 420w

“Literary merit aside, however, the authors all have a place in a

book which seeks not to present the best short stories but rather
different phases of American life. ‘American life of today,’ however, is
a misnomer. In their steadfast sometimes sentimental idealism, in
their passionate belief in democracy, the stories are obviously and
pathetically stories of life before the war.” Marian O’Connor
 + − N Y Evening Post p9 N 13 ’20 850w

“An unusually excellent anthology of American short tales.”

+ Outlook 126:201 S 29 ’20 120w

“Considered merely as a vehicle of recreational reading ‘Americans

all’ answers its purpose well; for the one who desires to combine
recreation with study of the successful short story the text is well
selected.”

+ Springf’d Republican p5a Ja 30 ’21

270w
Wis Lib Bul 16:194 N ’20 190w

HIBBEN, PAXTON. Constantine I and the Greek

people. il *$3.50 (3½c) Century 949.5

20–10649

The book was written in the spring of 1917 after the author had
been in Greece, Macedonia and Serbia and constitutes another
postwar revelation. It is stated that “during the war and after our
entry into it as an ally of France and Great Britain, without our
knowledge and consent the constitution of a little, but a brave and
fine people was nullified by the joint action of two of our allies: the
neutrality of a small country was violated, the will of its people set at
naught, its laws broken, its citizens persecuted, its press muzzled. By
force a government was imposed on this free people, and by force
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebooknice.com