Business Information Systems

Workshops: BIS 2019 International

Workshops, Seville, Spain, June 26–28,
2019, Revised Papers Witold
Abramowicz
Visit to download the full and correct content document:
https://textbookfull.com/product/business-information-systems-workshops-bis-2019-in
ternational-workshops-seville-spain-june-26-28-2019-revised-papers-witold-abramowi
cz/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Biota Grow 2C gather 2C cook Loucas

https://textbookfull.com/product/biota-grow-2c-gather-2c-cook-
loucas/

Business Information Systems: 22nd International

Conference, BIS 2019, Seville, Spain, June 26–28, 2019,
Proceedings, Part II Witold Abramowicz

https://textbookfull.com/product/business-information-
systems-22nd-international-conference-bis-2019-seville-spain-
june-26-28-2019-proceedings-part-ii-witold-abramowicz/

Business Information Systems: 22nd International

Conference, BIS 2019, Seville, Spain, June 26–28, 2019,
Proceedings, Part I Witold Abramowicz

https://textbookfull.com/product/business-information-
systems-22nd-international-conference-bis-2019-seville-spain-
june-26-28-2019-proceedings-part-i-witold-abramowicz/

Business Information Systems Workshops: BIS 2017

International Workshops, Pozna■, Poland, June 28-30,
2017, Revised Papers 1st Edition Witold Abramowicz
(Eds.)
https://textbookfull.com/product/business-information-systems-
workshops-bis-2017-international-workshops-poznan-poland-
june-28-30-2017-revised-papers-1st-edition-witold-abramowicz-eds/
Business Information Systems Workshops: BIS 2020
International Workshops, Colorado Springs, CO, USA,
June 8–10, 2020, Revised Selected Papers Witold
Abramowicz
https://textbookfull.com/product/business-information-systems-
workshops-bis-2020-international-workshops-colorado-springs-co-
usa-june-8-10-2020-revised-selected-papers-witold-abramowicz/

Business Information Systems Workshops BIS 2014

International Workshops Larnaca Cyprus May 22 23 2014
Revised Papers 1st Edition Witold Abramowicz

https://textbookfull.com/product/business-information-systems-
workshops-bis-2014-international-workshops-larnaca-cyprus-
may-22-23-2014-revised-papers-1st-edition-witold-abramowicz/

Business Information Systems Workshops: BIS 2016

International Workshops, Leipzig, Germany, July 6-8,
2016, Revised Papers 1st Edition Witold Abramowicz

https://textbookfull.com/product/business-information-systems-
workshops-bis-2016-international-workshops-leipzig-germany-
july-6-8-2016-revised-papers-1st-edition-witold-abramowicz/

Euro Par 2019 Parallel Processing Workshops Euro Par

2019 International Workshops Göttingen Germany August
26 30 2019 Revised Selected Papers Ulrich Schwardmann

https://textbookfull.com/product/euro-par-2019-parallel-
processing-workshops-euro-par-2019-international-workshops-
gottingen-germany-august-26-30-2019-revised-selected-papers-
ulrich-schwardmann/

Business Information Systems: 23rd International

Conference, BIS 2020, Colorado Springs, CO, USA, June
8–10, 2020, Proceedings Witold Abramowicz

https://textbookfull.com/product/business-information-
systems-23rd-international-conference-bis-2020-colorado-springs-
co-usa-june-8-10-2020-proceedings-witold-abramowicz/
 Witold Abramowicz
Rafael Corchuelo (Eds.)

Business
Information Systems
LNBIP 373

Workshops
BIS 2019 International Workshops
Seville, Spain, June 26–28, 2019
Revised Papers

123
Lecture Notes
in Business Information Processing 373

Series Editors
Wil van der Aalst
RWTH Aachen University, Aachen, Germany
John Mylopoulos
University of Trento, Trento, Italy
Michael Rosemann
Queensland University of Technology, Brisbane, QLD, Australia
Michael J. Shaw
University of Illinois, Urbana-Champaign, IL, USA
Clemens Szyperski
Microsoft Research, Redmond, WA, USA
More information about this series at http://www.springer.com/series/7911
Witold Abramowicz Rafael Corchuelo (Eds.)
•

Business
Information Systems
Workshops
BIS 2019 International Workshops
Seville, Spain, June 26–28, 2019
Revised Papers

123
Editors
Witold Abramowicz Rafael Corchuelo
Poznań University of Economics University of Seville
and Business Seville, Spain
Poznan, Poland

ISSN 1865-1348 ISSN 1865-1356 (electronic)

Lecture Notes in Business Information Processing
ISBN 978-3-030-36690-2 ISBN 978-3-030-36691-9 (eBook)
https://doi.org/10.1007/978-3-030-36691-9
© Springer Nature Switzerland AG 2019
The chapter “Competing for Amazon’s Buy Box: A Machine-Learning Approach” is Open Access. This
chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://
creativecommons.org/licenses/by/4.0/). For further details see license information in the chapter.
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, expressed or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
 Preface

In 2019 we had a great opportunity to organize the 22nd edition of the International
Conference on Business Information Systems (BIS 2019), that has grown to be a
well-renowned event for scientific and business communities. This year the main topic
of the conference was “Data Science for Business Information Systems.” The con-
ference was jointly organized by the University of Seville, Spain, and the Poznań
University of Economics and Business, Poland, and was held in Seville, Spain.
During each edition of the BIS conference series we make the effort to provide an
opportunity for discussion about up-to-date topics from the area of information systems
research. However, there are many topics that deserve particular attention. Thus, a
number of workshops and accompanying events are co-located with the BIS conference
series. The workshops give researchers the possibility to share preliminary ideas and
initial experimental results, and to discuss research hypotheses from a specific area of
interest.
Nine workshops and one accompanying event took place during BIS 2019. We were
pleased to host well-known workshops such as AKTB (11th edition), BITA (10th
edition), iCRM (4th edition), and iDEATE (4th edition), as well as relatively new
initiatives such as ISAMD, DigEx, BSCT, SciBOWater, and QOD. Each workshop
focused on a different topic: knowledge-based business information systems (AKTB),
challenges and current state of business and IT alignment (BITA), integrated social
CRM (iCRM), Big Data and business analytics ecosystems (iDEATE), Blockchain
(BSCT), digital customer experience (DigEx), maritime systems (ISAMD), water
management (SciBOWater), and data quality (QOD).
Additionally, BIS 2019 hosted a Doctoral Consortium. It was organized in a
workshop format, thus the best papers from this event are included in this book.
Moreover, all authors had the possibility to discuss their ideas on PhD thesis and
research work with a designated mentor.
The workshop authors had the chance to present their results and ideas in front of a
well-focused audience; thus the discussion gave the authors new perspectives and
directions for further research. Based on the feedback received, authors had the
opportunity to update their workshop articles for the current publication. This volume
contains 57 articles that are extended versions of papers accepted for BIS workshops.
In total, there were 139 submissions for all mentioned events. Based on the reviews, the
respective workshop chairs accepted 57 in total, yielding an acceptance rate of 41%.
We would like to express our thanks to everyone who made BIS 2019 workshops
successful. First of all, our workshops chairs, members of the workshop Program
vi Preface

Committees, authors of submitted papers, and finally all workshops participants. We

cordially invite you to visit the BIS website at https://bisconf.org/ and to join us at
future BIS conferences.

June 2019 Witold Abramowicz

Rafael Corchuelo
 Contents

AKTB Workshop

A Practical Grafting Model Based Explainable AI for Predicting

Corporate Financial Distress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Tsung-Nan Chou

Data Analytics in the Electronic Games . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Tomáš Porvazník, František Babič, and Ľudmila Pusztová

Evaluating the Interdependent Effect for Likert Scale Items . . . . . . . . . . . . . 26

Dalia Kriksciuniene, Virgilijus Sakalauskas, and Roman Lewandowski

Knowledge-Based UML Use Case Model Transformation Algorithm. . . . . . . 39

Ilona Veitaite and Audrius Lopata

Design of a Social-Based Recommendation Mechanism

for Peer-to-Peer Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Jyh-Hwa Liou, Ting-Kai Hwang, Sai-Nan Wu, and Yung-Ming Li

Mining Personal Service Processes: Towards a Conceptualization

for the Time Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Birger Lantow, Tom Baudis, and Fabienne Lambusch

Company Investment Recommendation Based on Data

Mining Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Svetla Boytcheva and Andrey Tagarev

BITA Workshop

An Exploration of Enterprise Architecture Research in Hospitals. . . . . . . . . . 89

Johannes Wichmann and Matthias Wißotzki

In Search for a Viable Smart Product Model . . . . . . . . . . . . . . . . . . . . . . . 101

João Barata and Paulo Rupino da Cunha

Strategic IT Alignment and Business Performance in SMEs:

An Empirical Investigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Fotis Kitsios and Maria Kamariotou

Enterprise Computing: A Case Study on Current Practices

in SAP Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Johannes Hintsch and Klaus Turowski
viii Contents

Integration of Enterprise Modeling and Ontology Engineering as Support

for Business/IT-Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Kurt Sandkuhl, Holger Lehmann, and Tom Sturm

Towards Aligning IT and Daily Routines of Older Adults . . . . . . . . . . . . . . 150

Marite Kirikova, Ella Kolkowska, Piotr Soja, Ewa Soja,
and Agneta Muceniece

Organizational Challenges of Digitalization Initiatives in Tourism Network

Management Organizations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Susanne Marx

A Configurational Approach to Task-Technology Fit

in the Healthcare Sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Patrick Mikalef and Hans Yngvar Torvatn

Ontology-Based Fragmented Company Knowledge Integration:

Multi-aspect Ontology Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Nikolay Shilov and Nikolay Teslya

BSCT Workshop

Comparing Market Phase Features for Cryptocurrency and Benchmark

Stock Index Using HMM and HSMM Filtering . . . . . . . . . . . . . . . . . . . . . 195
David Suda and Luke Spiteri

Contagion in Bitcoin Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

Célestin Coquidé, José Lages, and Dima L. Shepelyansky

Towards Blockchain and Semantic Web. . . . . . . . . . . . . . . . . . . . . . . . . . . 220

Juan Cano-Benito, Andrea Cimmino, and Raúl García-Castro

Detecting Brute-Force Attacks on Cryptocurrency Wallets . . . . . . . . . . . . . . 232

E. O. Kiktenko, M. A. Kudinov, and A. K. Fedorov

Analyzing Transaction Fees with Probabilistic Logic Programming . . . . . . . . 243

Damiano Azzolini, Fabrizio Riguzzi, and Evelina Lamma

An On-Chain Method for Automatic Entitlement Management

Using Blockchain Smart Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
Timothy Nugent, Fabio Petroni, Benedict Whittam Smith,
and Jochen L. Leidner

Study of Factors Related to Grin Cryptocurrency Mining Efficiency

with GPUs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Paulius Danielius, Tomas Savenas, and Saulius Masteika
 Contents ix

Towards Blockchain-Based E-Voting Systems . . . . . . . . . . . . . . . . . . . . . . 274

Chiara Braghin, Stelvio Cimato, Simone Raimondi Cominesi,
Ernesto Damiani, and Lara Mauri

Internet of Things and Blockchain Integration: Use Cases

and Implementation Challenges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Kelechi G. Eze, Cajetan M. Akujuobi, Matthew N. O. Sadiku,
Mohamed Chouikha, and Shumon Alam

Wikipedia as an Information Source on Cryptocurrency Technology . . . . . . . 299

Piotr Stolarski and Włodzimierz Lewoniewski

DigEX Workshop

Towards Analyzing High Street Customer Trajectories - A Data-Driven

Case Study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
C. Ingo Berendes

How Are Negative Customer Experiences Formed? A Qualitative Study

of Customers’ Online Shopping Journeys . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Tiina Kemppainen and Lauri Frank

A Model to Assess Customer Alignment Through Customer

Experience Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Leonardo Muñoz and Oscar Avila

Understanding Users’ Preferences for Privacy and Security Features –

A Conjoint Analysis of Cloud Storage Services . . . . . . . . . . . . . . . . . . . . . 352
Dana Naous and Christine Legner

The Role of Location Dependent Services for the Success of Local

Shopping Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Lars Bollweg, Richard Lackes, Markus Siepermann, and Peter Weber

iCRM Workshop

Social CRM Services in Digital Marketing Agencies: A Preliminary

Study on Service Offerings in Germany . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Julio Viana, Maarten van der Zandt, Olaf Reinhold, and Rainer Alt

Social Network Advertising Classification Based on Content Categories . . . . 396

Gustavo Nogueira de Sousa, Gustavo R. Almeida, and Fábio Lobato
x Contents

iDEATE Workshop

Developing an Artificial Intelligence Capability: A Theoretical

Framework for Business Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Patrick Mikalef, Siw Olsen Fjørtoft, and Hans Yngvar Torvatn

Measuring Qualitative Performance Criteria with Fuzzy Sets . . . . . . . . . . . . 417

Harry Martin

SmartM: A Non-intrusive Load Monitoring Platform . . . . . . . . . . . . . . . . . . 424

Xiufeng Liu, Simon Bolwig, and Per Sieverts Nielsen

Towards a Digitized Understanding of the Skilled Crafts Domain . . . . . . . . . 435

Maximilian Derouet, Deepak Nagaraj, Erik Schake, and Dirk Werth

Competing for Amazon’s Buy Box: A Machine-Learning Approach . . . . . . . 445

Álvaro Gómez-Losada and Néstor Duch-Brown

ISAMD Workshop

Spatial Query Processing on AIS Data Streams in Data Stream

Management Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
Tobias Brandt and Marco Grawunder

A Study of Vessel Trajectory Compression Based on Vector Data

Compression Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
Yuanyuan Ji, Wenhai Xu, and Ansheng Deng

OCULUS Sea™ Forensics: An Anomaly Detection Toolbox

for Maritime Surveillance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
Stelios C. A. Thomopoulos, Constantinos Rizogannis,
Konstantinos Georgios Thanos, Konstantinos Dimitros,
Konstantinos Panou, and Dimitris Zacharakis

Correcting the Destination Information in Automatic Identification

System Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
Matthias Steidel, Arne Lamm, Sebastian Feuerstack, and Axel Hahn

QOD Workshop

A New Tool for Automated Quality Control of Environmental Time

Series (AutoQC4Env) in Open Web Services . . . . . . . . . . . . . . . . . . . . . . . 513
Najmeh Kaffashzadeh, Felix Kleinert, and Martin G. Schultz
 Contents xi

Approach to Improving the Quality of Open Data in the Universe

of Small Molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
John L. Markley, Hesam Dashti, Jonathan R. Wedell,
William M. Westler, Eldon L. Ulrich, and Hamid R. Eghbalnia

Evaluating the Quantity of Incident-Related Information

in an Open Cyber Security Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
Benjamin Aziz, John Arthur Lee, and Gulsum Akkuzu

Semantic Data Integration and Quality Assurance of Thematic Maps

in the German Federal Agency for Cartography and Geodesy . . . . . . . . . . . . 543
Timo Homburg, Sebastian Steppan, and Falk Würriehausen

Technical Usability of Wikidata’s Linked Data: Evaluation

of Machine Interoperability and Data Interpretability . . . . . . . . . . . . . . . . . . 556
Nuno Freire and Antoine Isaac

SciBOWater Workshop

Telemetry System for Smart Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . 573

C. M. Balaceanu, I. Marcu, and G. Suciu

Increasing Collaboration and Participation Through Serious Gaming

for Improving the Quality of Service in Urban Water Infrastructure . . . . . . . . 585
Alexandru Predescu and Mariana Mocanu

Information Technology for Ethical Use of Water . . . . . . . . . . . . . . . . . . . . 597

Panagiotis Christias and Mariana Mocanu

Doctoral Consortium

Towards a System for Data Transparency to Support Data Subjects. . . . . . . . 613

Christian Janßen

Towards a Record Linkage Layer to Support Big Data Integration . . . . . . . . 625

Felix Kruse

Incremental Modeling of Supply Chain to Improve Performance Measures. . . 637

Szczepan Górtowski and Elżbieta Lewańska

Use of Data Science for Promotion Optimization in Convenience Chain . . . . 649

Sławomir Mazurowski and Elżbieta Lewańska

Towards a Cross-Company Data and Model Platform for SMEs . . . . . . . . . . 661

René Kessler
xii Contents

Touchscreen Behavioural Biometrics Authentication in Self-contained

Mobile Applications Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672
Piotr Kałużny

Data-Based User’s Personality in Personalizing Smart Services . . . . . . . . . . . 686

Izabella Krzeminska

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697

AKTB Workshop
 AKTB 2019 Workshop Chairs’ Message

AKTB 2019 was the 11th Workshop on Applications of Knowledge-Based Tech-

nologies in Business, organized in conjunction with the BIS conference series. It
continued the successful series of AKTB workshops in Poznan, Berlin, Vilnius,
Larnaca, Leipzig, etc. AKTB 2019 invited researchers, practitioners, and policy makers
to share knowledge of efficient computational intelligence methods for implementing
business information systems in finance, healthcare, e-business, and other application
domains. We requested original papers which could provide advanced services based
on innovative data science approaches for the information systems users, and propose
efficient solutions for smart business and process modeling. Authors from 15 countries
submitted 22 articles to the AKTB 2019 workshop. Each paper was evaluated by two
or three independent reviewers of the Program Committee (PC). The highest ranked ten
articles, representing the research trend from nine different countries, were accepted for
presentation during the conference and the second stage of reviewing before including
them into the post-conference proceedings. The PC of AKTB 2019 invited one addi-
tional paper disclosing the evaluation of the interdependent effect for Likert scale items.
The 21 outstanding researchers who represent prestigious scientific institutions from 8
countries joined the PC as paper reviewers. They evaluated the quality of the articles by
taking into account the criteria of its relevance to the workshop topics, originality,
novelty, and quality of presentation. We appreciate the level of work and expertise of
the PC members, whose reviews provided deep analysis of the submitted research
works and highlighted valuable insights for the authors. High standards followed by
reviewers ensured a high quality workshop event, excellent presentations, intensive
scientific discussions, and added value to the post-conference workshop proceedings.
We would like to express our gratitude for the joint input of many people who made
AKTB 2019 a success: to all authors of submitted papers, members of the PC, the
Kaunas Faculty of Vilnius University, Department of Information Systems of the
Poznan University of Economics, ETSI Informática, the University of Seville, and to
the outstanding efforts of the Organizing Committee of the 22nd International Con-
ference BIS 2019.

Virgilijus Sakalauskas
Dalia Kriksciuniene
 Organization

Chairs

Virgilijus Sakalauskas Vilnius University, Lithuania

Dalia Kriksciuniene Vilnius University, Lithuania

Program Committee

María Dolores Afonso Suárez University of Las Palmas de Gran Canaria,

Spain
Lia Bassa Foundation for Information Society, Spain
Dumitru Dan Burdescu University of Craiova, Romania
Ferenc Kiss Budapest University of Technology
and Economics, Hungary
Dalia Kriksciuniene Vilnius University, Lithuania
Dariusz Krol Wrocław University of Science
and Technology, Poland
Roman Lewandowski University of Social Sciences, Poland
Audrius Lopata Vilnius University, Lithuania
Dale Luksaite Kauno Kolegija University of Applied
Sciences, Lithuania
Saulius Masteika Vilnius University, Lithuania
Laima Papreckiene Kaunas University of Technology, Lithuania
Justyna Patalas-Maliszewska University of Zielona Góra, Poland,
and University of Vienna, Austria
Tomas Pitner Masaryk University, Czech Republic
Giedrius Romeika Vilnius University, Lithuania
José Raúl Romero University of Cordoba, Spain
Vytautas Rudzionis Vilnius University, Lithuania
Virgilijus Sakalauskas Vilnius University, Lithuania
Darijus Strasunskas HEMIT, Norway
Ilona Veitaite Vilnius University, Lithuania
Sebastián Ventura University of Cordoba, Spain
Danuta Zakrzewska Institute of Information Technology Technical
University of Lodz, Poland
 A Practical Grafting Model Based
Explainable AI for Predicting Corporate
Financial Distress

Tsung-Nan Chou(&)

Chaoyang University of Technology, Taichung 41349, Taiwan

[email protected]

Abstract. Machine learning and deep learning are all part of artificial intelli-
gence and have a great impact on marketing and consumers around the world.
However, the deep learning algorithms developed from the neural network are
normally regarded as a black box because their network structure and weights
are unable to be interpreted by a human user. In general, customers in the
banking industry have the rights to know why their applications have been
rejected by the decisions made by black box algorithms. In this paper, a practical
grafting method was proposed to combine the global and the local models into a
hybrid model for explainable AI. Two decision tree-based models were used as
the global models because their highly explainable ability could work as a
skeleton or blueprint for the hybrid model. Another two models including the
deep neural network and the k-nearest neighbor model were employed as the
local models to improve accuracy and interpretability respectively. A financial
distress prediction system was implemented to evaluate the performance of the
hybrid model and the effectiveness of the proposed grafting method. The
experiment results suggested the hybrid model based on the terminal node
grafting might increase the accuracy and interpretability depending on the
chosen local models.

Keywords: Decision tree grafting
Explainable AI
Deep neural network

1 Introduction

The machine learning and artificial intelligence (AI) technology have been successfully
applied to many financial services to solve various decision-making problems. How-
ever, the pervasive deep learning techniques developed from the neural network is
considered as a black box and cannot be explained by humans because their network
architecture and weights are difficult to be interpreted. Consequently, their processed
results are unable to confirm whether the decision-making and analysis are reasonable.
In addition, in the case of a consumer loan application, if the financial institution rejects
an applicant’s loan, the borrower has the rights to ask for the reason why the loan is not
approved. In such a situation, the credit decision generated by the automated decision
system using the deep learning algorithm needs to explain how the judgment process
work. In order to avoid possible irreparable harm from the automated decision-making
systems, the General Data Protection Regulation (GDPR) approved by the European

© Springer Nature Switzerland AG 2019

W. Abramowicz and R. Corchuelo (Eds.): BIS 2019 Workshops, LNBIP 373, pp. 5–15, 2019.
https://doi.org/10.1007/978-3-030-36691-9_1
6 T.-N. Chou

Union in 2016 and enforced in 2018 stipulates that rejected applicants have the rights to
know the details of the process leading to their rejection [1]. Therefore, for the most
common applications of machine learning in finance, legal consultation, and medical
diagnosis, the user must be informed with the reasons for decision or judgment.
Otherwise, even the decision is correct, or the prediction model is highly accurate, the
result is still doubtful.
Unlike the difficulty of understanding why artificial intelligence has failed, humans
can understand and discover their decision and judgment errors and then correct them
to avoid making the same mistakes. Therefore, many researchers have tried to develop
different methods to make the black boxes of artificial intelligence understandable
without affecting their efficiency and performance. As a result, the research area such as
the explainable AI (XAI), which is concerned with how to make humans understand
the reasons for judgment under artificial intelligence, is getting more attention. How-
ever, while using XAI models for prediction, these models need to deal with the
tradeoff between model interpretability and accuracy [2]. Generally, an effective XAI
model could be used to explain the decision rules generated, but the performance of the
model might be affected. Since the artificial intelligence is originally designed to solve
a large amount of data analysis problems that humans are difficult to handle, whether
the AI models can be understood is closely related to the complexity of the algorithms
they used. In general, the more complex a model is, the more difficult to interpret its
decision-making and reasoning processes. How to make artificial intelligence inter-
pretable without affecting its performance for an automated decision-making system
will be a challenge for researchers. In deep learning techniques, the generative
adversarial network (GAN) could be used to train two neural networks against each
other to maximize learning outcomes and explore the internal workings of neural
networks to provide interpretability [3]. However, the training of GAN is time-
consuming and might be unstable. In contrast, the decision tree generates a set of
decision rules that are easy for the human to interpret, and the algorithm could be one
of the solutions that help the banking industry to be compliant with the GDPR
requirements. Although the concepts of explainability and interpretability are different
for XAI applications, both are used interchangeably in this paper because the applied
decision trees are interpretable models that also provide understandability to know how
the results are achieved. The following sections introduce the methods and strategies
used in this study briefly.

2 Research Methods and Experiment Design

Some machine learning algorithms such as decision trees, Bayesian inference or Sparse
linear models (SLIM) [4] intrinsically provide human interpretability that is easy to
understand. Alternatively, many research approaches attempt to realize how the
machine learning model work by examining their global or local interpretability, and
some others use model-specific or model-agnostic to distinguish the way in which
interpretability can be generated. In general, model-agnostic approaches such as partial
dependence plot (PDP), Shapley value and local interpretable model-agnostic expla-
nations (LIME) provide interpretability that is compatible with any classification,
 A Practical Grafting Model Based Explainable AI 7

regression, or reinforcement learning models [5]. The PDP model [6] generally changes
the value of a certain characteristic variable one by one while controlling other vari-
ables, and interprets the relationship between the characteristic variable and the target
variable by a line graph. Comparatively, Lundberg and Lee [7] proposed the Shapely
explanations method, using the Shapley value designed by Lloyd Shapley [8] to
explain the importance of feature variables. In general, the deep learning models use the
training sample to learn the relationship between the input and output variables. If the
sample data increases, the mapping of relationship becomes more complicated and
makes the constructed model difficult to understand. The LIME algorithm uses a small
number of regional samples to construct a simple local model, and the importance of
the variable with respect to a testing sample is calculated by the local model rather than
the global model previously trained. In brief, this algorithm uses the local model as a
proxy model to interpret the original black-box global model [9]. Since the LIME
algorithm is model-agnostic, the global model can be any machine learning or deep
learning algorithm, while the local model can apply a simple linear regression or a
decision tree to fit the target values predicted by the original global model.
This paper proposed a practical grafting method with intent to overcome the
tradeoff between accuracy and interpretability in machine learning based on the inte-
gration of the global model and the local model. In the first place, each of five common
machine-learning models was applied as a global model to evaluate its individual
performance. Especially, the decision tree (DT) and fuzzy decision tree (FDT) were
used as the global modes due to their explainable ability. Both models worked as a
skeleton or blueprint to concatenate the local models. To balance the tradeoff between
accuracy and interpretability, the deep neural network (DNN) and k-nearest neighbor
methods (KNN) were chosen as local models, where the former intended to increase
the accuracy and the latter attempted to increase the interpretability. Finally, the ter-
minal node grafting method that integrated the global and the local models based on
three fusion strategies were implemented to examine the performance of the hybrid
model. The methods and strategies used in this study were briefly described as
following.

2.1 Interpretable Global and Local Models

According to the leading auditing and consulting company PwC’s subjective ranking
for the explainability of machine learning techniques [10], the decision tree acquired
the highest score among all compared models, while the neural network attained the
lowest score. Therefore, both the decision tree (DT) and the fuzzy decision tree
(FDT) were chosen as the global models due to their highly explainable ability. In
contrast to the DT model, the FDT model was employed as a counterpart to examine
whether the fuzzy membership and inference mechanism could improve the model
performance. The k-means algorithm was also used to generate optimized fuzzy
membership partitions for the input variables [11, 12]. Moreover, the random forest
(RF) model was another compared model since it randomly chose variables and
samples to build several decision trees and combined the outcomes of multiple decision
trees into a single average result. Apart from the tree-based models, the KNN and DNN
models were applied not only as an independent model to compare their performance,
8 T.-N. Chou

but also as a local model in conjunction with the DT or the FDT global model to
construct a hybrid system. The number of hidden layers and the corresponding pro-
cessing units within the sequential DNN model [13] were examined with different
experimental configurations. Furthermore, the dropout layers with a cut-off value of 0.1
were used to control the overfitting problem. The activation functions such as Sigmoid,
Tanh and ReLU were assessed in different hidden layers and another SoftMax function
was employed in the last fully connected layer.

2.2 The Decision Tree Grafting

By analogy with grafting in plants, heterogeneous models could concatenate on the
terminal or internal nodes of a decision tree that worked as the root system. For the
grafting performed on internal trunks or branches of a decision tree, the structure of the
original tree was altered and the new hybrid tree became more complex. However, if
the grafting performed on the terminal nodes, the structure of the original tree remained
unchanged. The terminal node grafting might at least not affect the interpretability of
the original decision tree and night also improve the model accuracy depending on
what kind of local model applied. Some research work [14] have demonstrated that the
combination of both pruning and grafting might effectively provide the best general
predictive accuracy. Although most of the grafting approaches added the new branches
with another decision tree to increase accuracy, we proposed a practical grafting
method by using the DNN and the KNN models as the local models rather than using
the tree-based model. Both the terminal node grafting and the internal node grafting
were developed to improve the accuracy and interpretability of the hybrid model.

Fig. 1. The terminal node grafting

 A Practical Grafting Model Based Explainable AI 9

As shown in Fig. 1, for any decision path ended with a terminal node, the subset of
training data partitioned by this path was collected to train another specified local
model if it performed poor accuracy. The model interpretability remained unchanged if
using the inexplicable DNN model as a local model, while using the KNN model might
increase the model interpretability. The other grafting method was the internal node
grafting illustrated in Fig. 2. The internal nodes (or decision nodes) of a decision tree
performing lower accuracy were removed and replaced with a local model such as the
DNN or the KNN model. The branches of a decision tree resembled the clustering
function to partition the training data into smaller subsets for training the local models.

Fig. 2. The internal node grafting

2.3 The Model Integration Approaches

The strength of the FDT model, with respect to other machine learning models, was its
inductive learning and linguistic rule construction that inherited from the decision tree
and fuzzy inference. Some applications such as medical diagnosis benefited greatly
from its interpretative power based on the readable rules. As the global model was the
root system and implemented as the skeleton or blueprint of a decision-making
framework, the model required to provide the explainable ability to help people
understand and trust its outcomes. Therefore, both the DT model and the FDT model
were selected as global models because of their interpretable decision rules and the
ability to handle inexact and uncertain information. The tree structures of both models
were converted to a set of if-then decision rules with a slight difference at multiple
antecedent statements. In contrast to the DT model, the FDT model applied the k-
means algorithm to partition all input variables into three fuzzy subsets denoted as
10 T.-N. Chou

MF1, MF2, and MF3. The decision rules generated by the DT and the FDT models
comprised 6 and 10 rules, respectively. As the excerpted rules illustrated in Fig. 3, the
fuzzy rules derived from the FDT model were controlled by the maximum of four input
variables in the non-crisp antecedents to reduce the complexity of the global model.

Fig. 3. The excerpted rules of the DT and the FDT models

Generally, the DT model searched for a single suitable if-then rule from its trained
rule set and applied it to the prediction of a testing sample. In contrast, the FDT model
required to aggregate multiple fuzzy rules to evaluate the overall effect on that pre-
diction. Therefore, this difference raised two challenges for implementing the FDT
model as a global model in grafting. The first challenge concerned about how to use the
multiple rules of the FDT model to train each of their corresponding local models. In
terminal node grafting, the training data that complied with a single decision rule of the
DT model was directly partitioned as an individual subset for training the local model.
In contrast, implementing the FDT model as a global model required to combine
several subsets of training data for the local models according to the aggregation of
multiple fuzzy rules. The other challenge was how to integrate the global model with
the local model to achieve the best synthesis results. If the FDT model was used as a
global model, each of its activated fuzzy rules imposed a different weighting effect on
the prediction of the testing sample.
Accordingly, three fusion methods were developed to quantify the effect of fuzzy
rules on their concatenated local models and calculate the synthetic outcome for the
testing data. The first fusion method was a weighted sum approach. The evaluation of a
testing sample was calculated by the weighted sum of the outputs from its corre-
sponding local models, where the weights were the relative importance derived from
the activated fuzzy rule associated with the local model. In addition, the second fusion
method evaluated the testing sample based on the average output of the local models
triggered by their corresponding activated fuzzy rules from the global model. More-
over, because each of the activated fuzzy rules represented a weighted vote to its
associated local model, the evaluation of a testing sample in the third fusion method
was decided by the activated fuzzy rule that acquired the relative majority. In this
paper, the FDT model implemented with the KNN model based on three fusion
methods were denoted as the FDT-KNN (1) to FDT-KNN (3) models. Comparatively,
 A Practical Grafting Model Based Explainable AI 11

the FDT model integrated with the DNN model were denoted as the FDT-DNN (1) to
FDT-DNN (3) models.

3 Experiment Results

To evaluate the integrated performance of the global and the local modes based on
different grafting and fusion methods, a financial distress prediction system was
implemented with various experimental configurations. The dataset for experiments
comprised the financial and corporate information of 32 distress and another 32 healthy
companies declared by the authoritative Taiwan Economic Journal (TEJ) in accordance
with various criteria and definitions during the period of 2008 and 2015. Furthermore,
the dataset consisted of 15 independent variables were subdivided into 80% for the
training dataset and another 20% for the testing dataset. The DT, FDT and KNN
models were all regarded as interpretable models regardless of whether they were used
as the global or local model. Totally, five experiments were established to evaluate the
performance of the model grafting strategies based on four metrics including accuracy,
kappa, sensitivity, and specificity. Considering the tradeoff between the accuracy and
interpretability for training tree-based global models, the complexity parameter
(CP) controlling the number of splits in a decision tree by examining the misclassifi-
cation error for each branch was evaluated with 0.01 and 0.001, and both were denoted
as the DT1 and DT2 models in the first experiment. Moreover, the minimum number of
samples in a terminal node and the maximum depth and height of a tree were also
controlled to construct a concise global model. As shown in Table 1, the experiment
result reported that the DT2 model demonstrated better training accuracy of 0.89, only
5% difference from the simpler DT1 model. However, the DT2 model created a larger
rule set of 12 if compared to the smaller DT1 model generating 6 rules only. Although
the accuracy of DT1 model was slightly lower than the DT2 model, the DT1 model was
selected as the global model to concatenate local models because it was easier to
interpret, and the local models might recover the minor decrease of accuracy during
grafting experiments.

Table 1. The DT model trained with different CP Parameters.

Trained model Accuracy Kappa Sensitivity Specificity
DT1 (CP = 0.01) 0.84 0.67 0.86 0.81
DT2 (CP = 0.001) 0.89 0.78 0.91 0.87

The second experiment intended to compare the performance of the five common
models described in Sect. 2 prior to working as either a local model or global model.
As indicated in Table 2, the FDT model was inferior to the other four models although
all models achieved the accuracy over 0.70. The accuracy of the FDT model was 0.73,
a 6% deteriorated from the original DT model being reported. The reason why the DT
model performed better than the FDT model was that the latter was pruned to preserve
a similar tree structure as its counterpart DT model for further comparison of predictive
12 T.-N. Chou

accuracy after grafting, and the crisp inputs needed to be transformed into fuzzy inputs
for a fuzzy inference rule. As the random forest was an ensemble approach that
combined multiple decision trees into a single predictive model to reduce variance and
bias, the experiment result unquestionably suggested the RF model was the best model
and achieved an accuracy of 0.85, which outperformed all other common models.

Table 2. The performance comparison of five machine-learning models

Model Accuracy Kappa Sensitivity Specificity
DT 0.79 0.58 0.88 0.71
FDT 0.73 0.48 0.94 0.56
RF 0.85 0.70 0.94 0.77
KNN 0.80 0.60 0.94 0.67
DNN 0.81 0.63 0.90 0.74

The model grafting created a hybrid model that combined the global model with the
local model to increase the overall interpretability and accuracy. Both the DT and the
FDT models could be employed as a global mode because they provided better
interpretability for the human to understand. On the other hand, the KNN model used
as a local model with an attempt to increase more interpretability, while the DNN
model was applied to improve predictive accuracy. The performance of the terminal
node grafting based on using the DT model as a global model was reported in Table 3.
As opposed to the single DT model obtained the accuracy of 0.79, the DT-based
grafting result indicated a moderate improvement, and the accuracy was increased to
0.87 for concatenated with the DNN model and 0.82 for the KNN model respectively.
Although the DT-DNN model achieved higher accuracy, however, the DT-KNN model
gained more interpretability by slightly losing the accuracy of 5%.

Table 3. The grafting performance of the DT-DNN and DT-KNN models

Model Accuracy Kappa Sensitivity Specificity
DT-DNN 0.87 0.73 0.88 0.85
DT-KNN 0.82 0.65 0.96 0.70

Since the FDT model required to concatenate several local models together due to
the multiple fuzzy rules been activated simultaneously, three fusion methods were
applied to merge the outcomes of the FDT model with its corresponding local models.
By referring to Table 4, the accuracy of the FDT-DNN (3) model using the third fusion
method performed the lowest accuracy of 0.73, the same as the result of using the FDT
model alone. On the other hand, the FDT-DNN (1) model using the first fusion method
improved the accuracy to 0.88 and it was considered as the best model if compared to
the other two fusion methods.
 A Practical Grafting Model Based Explainable AI 13

Table 4. The grafting performance of the FDT-DNN models

Model Accuracy Kappa Sensitivity Specificity
FDT-DNN (1) 0.88 0.75 0.92 0.84
FDT-DNN (2) 0.82 0.65 0.9 0.75
FDT-DNN (3) 0.73 0.46 0.75 0.72

In the final experiment, the KNN model was employed as a local model in con-
junction with the FDT model to evaluate three fusion methods. As reported in Table 5,
the FDT-KNN (1) model outperformed the other models and achieved an accuracy of
0.84. However, this result was slightly lower than that of the FDT-DNN (1) model
using the same fusion method and suggested that the KNN model might not be able to
improve the predictive accuracy substantially. Although the DNN model aided the
hybrid model to improve the predictive accuracy, the KNN model implicated more
explanations could be obtained to learn how the results were achieved.

Table 5. The grafting performance of the FDT-KNN models

Model Accuracy Kappa Sensitivity Specificity
FDT-KNN (1) 0.84 0.68 0.98 0.72
FDT-KNN (2) 0.81 0.63 0.94 0.71
FDT-KNN (3) 0.76 0.51 0.69 0.82

As a highlight, while using only a single model in the experiment, the RF model
achieved the highest predictive accuracy among five testing models. However, the
interpretability of the RF model was limited since the model applied an ensemble of a
number of decision trees. Moreover, by using a hybrid model that integrated a global
model with several local models, the predictive results varied and depended on which
fusion method was used in model grafting. As the experiment results suggested, the
first fusion method surpassed the other two fusion methods, and the combination of the
FDT and DNN models might outperform the single RF model in predictive accuracy.
Conversely, by using the FDT model in conjunction with the KNN model, the pre-
dictive accuracy was slightly worse than the RF model. If considering the DT model
instead of the FDT model to concatenate with the local models, the combination of the
DT and the DNN model also performed better than the single RF model. Consequently,
both the DT and FDT model grafted with the DNN model might be the favorable
hybrid solutions to increase predictive accuracy, even by losing a bit of interpretability
if compared with the KNN model.

4 Conclusion

To create a fully interpretable framework required combining an explainable global

model with flexibly selected local models to prevent from losing accuracy or inter-
pretability. In this paper, a practical grafting approach was proposed to integrate the
14 T.-N. Chou

global model and the local model together with the intent to improve the overall
interpretability and accuracy of the resulting hybrid model. Both the DT and the FDT
models were chosen as the global models due to their highly explainable ability. On the
other hand, the inexplicable DNN model and the explainable KNN model were applied
as local models to improve the accuracy and interpretability respectively. By
employing a signal model in the experiment, the result indicated the RF model was the
best model with the predictive accuracy of 0.85. However, if we implemented a hybrid
model based on three fusion methods in the terminal node grafting, the FDT-DNN
(1) model was considered as the most competent model under all experiments and
achieved the highest accuracy of 0.88. In addition, the experiment result also suggested
that both the DT and the FDT models were capable of providing the explainable ability
and could be combined with the DNN model to improve model performance by using
the proposed grafting technique. Although the KNN model might deliver additional
interpretability to the hybrid model, the predictive accuracy was slightly decreased.
This research work can be further extended to implement the internal node grafting and
compare its performance with that of terminal node grafting described in this study.

References
1. Goodman, B., Flaxman, S.: European union regulations on algorithmic decision-making and
a “right to explanation”. AI Mag. 38, 50–57 (2017)
2. Adadi, A., Berrada, M.: Peeking inside the black-box: a survey on explainable artificial
intelligence (XAI). IEEE Access 6, 52138–52160 (2018)
3. Karras, T., Aila, T., Laine, S., Lehtinen, J.: Progressive growing of GANs for improved
quality, stability, and variation. arXiv:1710.10196v3 [cs.ne] (2018)
4. Ustun, B., Rudin, C.: Supersparse linear integer models for optimized medical scoring
systems. Mach. Learn. 102(3), 349–391 (2016)
5. Došilović, F.K., Brčić, M., Hlupić, N.: Explainable artificial intelligence: a survey. In: 2018
41st International Convention on Information and Communication Technology, Electronics
and Microelectronics (MIPRO), pp. 0210–0215 (2018)
6. Berk, R., Bleich, J.: Statistical procedures for forecasting criminal behavior: a comparative
assessment. Criminol. Public Policy 12(3), 513–544 (2013)
7. Lundberg, S.M., Lee, S.I.: A unified approach to interpreting model predictions. In:
Advances in Neural Information Processing Systems (NIPS 2017), vol. 30, pp. 4765–4774
(2017)
8. Roth, A.E. (ed.): The Shapley Value, Essays in Honor of Lloyd S. Shapley. Cambridge
University Press, Cambridge (1988)
9. Ribeiro, M.T., Singh, S., Guestrin, C.: Why should I trust you? Explaining the predictions of
any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining, pp. 1135–1144 (2016)
10. Rao, A., Golbin, I.: What it means to open AI’s black box. Next in Tech, PwC Blogs. http://
usblogs.pwc.com/emerging-technology/to-open-ai-black-box (2018), last accessed 2019/
04/14
11. Levashenko, V., Zaitseva, E.: Fuzzy decision trees in medical decision making support
system. In: Proceedings of 2012 Federated Conference on Computer Science and
Information Systems, pp. 213–219 (2012)
 A Practical Grafting Model Based Explainable AI 15

12. Guillaume, S., Charnomordic, B.: Learning interpretable fuzzy inference systems with
Fispro. Int. J. Inform. Sci. 181(20), 4409–4427 (2011)
13. Liu, W., Wang, Z., Liu, X., Zeng, N., Liu, Y., Fuad, E.A.: A survey of deep neural network
architectures and their applications. Neurocomputing 234, 11–26 (2017)
14. Webb, G.I.: Decision tree grafting from the all tests-but-one partition. In: Proceedings of the
16th International Joint Conference on Artificial Intelligence (IJCAI), vol. 2, pp. 702–707
(1999)
 Data Analytics in the Electronic Games

Tomáš Porvazník, František Babič(&), and Ľudmila Pusztová

Department of Cybernetics and Artificial Intelligence,

Faculty of Electrical Engineering and Informatics,
Technical University of Košice, 040 01 Košice, Slovak Republic
[email protected], [email protected],
[email protected]

Abstract. This paper aims at the use of data analytics methods in mobile
games. The main goal was to predict future purchases of players in the selected
mobile game. The result presents the information about whether the player is
going to buy any of the offered bonus packages or not. This information is
crucial for marketing and possible ways of monetization. From the perspective
of data analytics, the goal is the creation of a classification model in line with the
CRISP-DM methodology. We used the following algorithms in the modeling
phase: Random forest, Naive Bayes, Linear regression, XGBoost, and Gradient
Boosting. All generated models were evaluated by contingency tables, which
presented models accuracy as the ration between successfully predicted values
to all predicted samples. The results are plausible and have the potential to be
deployed into practice as a baseline model or support for personalized marketing
activities.

Keywords: Mobile games
Data analytics
Monetization

1 Introduction

Everybody knows what mobile games are. We play them in our free time, on the bus,
some of us even at work. Mobile games sector is a highly competitive, and the
development of a successful game is a challenging and complex issue. A few years ago,
you came to the shop and bought a physical copy of the game. Nowadays, the process
is even more straightforward with the availability of online mobile stores. But the
players prefer another model for mobile games; it is called a fremium. The 2018 report
by Nielsen gaming research company SuperData has revealed that so-called freemium
games – games that are free to play but typically enable microtransactions for users to
purchase upgrades and gizmos – are dominating the entertainment market, generating a
massive $88 billion in the past year. This fact motivated us to start a collaboration with
one of the game development companies creating free-to-play games. During the
analytical process, we communicated intensively with the experts; not only during the
data understanding but also in the evaluation phase.
The paper is organized as follows: an introduction with motivation, related work,
methodology, and used methods. The description of the performed analytical project is
in line with the CRISP-DM methodology. The conclusion summarizes the results.

© Springer Nature Switzerland AG 2019

W. Abramowicz and R. Corchuelo (Eds.): BIS 2019 Workshops, LNBIP 373, pp. 16–25, 2019.
https://doi.org/10.1007/978-3-030-36691-9_2
 Data Analytics in the Electronic Games 17

1.1 Electronic Games

Field of electronic games is growing every day. Generally speaking, it is a game played
on some electronic device, whether it is a console, mobile phone or computer. Some of
the most popular mobile games count their daily active users in millions. The Q4 and
Full Year 2018 Store Intelligence Data Digest report1 pointed out that worldwide app
revenue grew 22.7% in 2018 to $71.3 billion and new app installs topped 105.3 billion,
increasing 11.1 percent over 2017. In the mobile games world, the hyper-casual titles
dominated the download charts for 2018, led by releases from Voodoo (Helix Jump)
and Lion Studios (Love Balls).
If the mobile game uses the freemium model, the idea is that they make a lot of
money. But, they have to have a good monetization strategy. The recommended way
how to do this effectively is to follow the ARM model invented by Lewis in 1898 [1].
ARM stands for Acquisition, Retention, and Monetization. You may say that mobile
games did not even exist back then. That is because this model is not focused on mobile
games but business in general. Gain new customers, make them stay and in the end,
monetize them. In each step, you have to define relevant key performance indicators [2].
The first step in the model is an acquisition. Use every available source and gain as
many players as possible. Viral sources are better than those nonviral because nonviral
sources cost studio money. Once your game went viral, you are all good. K-factor is a
metric representing several users who come from virality (social media shares). The
second kind of acquisition, nonviral, is based on ads and marketing. There are few
indicators if your ad campaign is successful. One of them is Cost per Install. This key
performance indicator measures how much you need to spend on ads to gain one user.
The second step in the ARM model is retention. Longer, the user will stay the
higher profit you will have from him. If the user is coming back to the game regularly,
there is a bigger chance of monetizing him. For example, you can you two approaches
to motivate him: reward him for coming back or punish him for not doing so. The
punishment method is outdated. The modern game studios are using rewarding even
for failure in the game. It is all mind games. In this step, there are also a few essential
key performance indicators like daily active users, monthly active users, and an average
total in-app time. Let’s say you have gain users and they are coming back every so
often.
The next step will be to monetize the users. In-app purchases and ads are the only
ways of profiting in the freemium model. This step aims to convert as many players to
paying ones as possible. Not every player is willing to spend money in the game. The
fact is that paying users to represent only a fraction of all players, mostly less than 5%.
Those paying users are commonly separated into three categories. First, there are
minnows. The players, who spend a little bit of money, maximum of 5$ in their lifetime
(total time spend in the game). Next, the dolphins, those spend up to 20$. And the
smallest group but with the highest earnings is called whales. Those players spend
more than 20$, sometimes even more than 100$. The game must be designed in such a
way that it will encourage users to pay. To monitor monetization there are several key

1
https://sensortower.com/blog/top-apps-games-publishers-2018.
Another random document with
no related content on Scribd:
 Nestes dias festivaes
Com summa gala e grandeza,
Assistiu toda a nobreza
Dos homens mais principaes,
Ministros, officiaes
De guerra e damas mui bellas,
Que em palanques e janellas
Mostravam com arrebol,
Que estando alli posto o sol,[5]
Bem podiam ser estrellas.

[5] Refere-se ao governador.

Posto o sol alli se via,

Porém com notavel gosto,
Quando vi que era o sol posto
Mais o Terreiro luzia:
Dois soes[6] postos na Bahia
Vi com differença atroz
Um Saturno que se poz,
Outro posto na janella
Sol de luz mais clara e bella,
Que hoje nasce para nós.

[6] Allude aos dois governadores d. J. de Aloncastre e Camara

Coutinho, que se achavam presentes.

Desterrando sombras mil

De um sol que causou desmaios,
Nasce com benignos raios
Este sol para o Brazil:
Oh quem tivera a subtil
De Apollo lyra discreta,
Da fama a aguda trombeta,
Para que pudesse ousado,
Sem temor, nem perturbado
Descrever este planeta.

Mas é fraco o meu engenho

Para de um sol sem desmaios
Querer ventilar os raios
Quando olhos de aguia não tenho:
E si a tão sublime empenho,

Onde o mais sabio delira,

Meu pensamento subira,
Logo d’essa esphera clara
Como Phaetonte rodára,
Ou como Icaro cahira.

Quando o planeta maior

Á vista humana se expõe,
É que a seus raios se oppõe
Atrevido algum vapor:
E si neste sol melhor
Nenhuns eclypses se vêm,
Não se atreverá ninguem
Sem ter de nescio desmaios
Querer contemplar os raios
Esclarecidos que tem.

Quando da esteril mulher

Nasceu o maior do mundo
Admirações e profundo
Pasmo veiu a gente a ter:
E si com João nascer
Houve tanta admiração,
Á Bahia outro João,
Sol de claro nascimento
Nasce com merecimento
Para a mesma suspensão.
E como não pasmarei
Eu, e este povo tambem,
De ter por General, quem
Sceptro merece de rei?
Pois a ventura e a lei
Divina dispoz, senhor,
O seres Governador,
Com tudo sabemos nós,
Que um foi de vossos avós
De Pedro progenitor.

D’aquelle em tudo primeiro

João era nada segundo
Sois, e bem conhece o mundo,
Descendente verdadeiro:
Tambem da Casa de Aveiro
Muita nobreza alcançaes:
Alencastre vos chamaes,
De Duarte Inglez potente
Clarissimo descendente:
Silva sois, não digo mais.

Com branca e encarnada pluma

Galan vestido de verde,
Que inda a esperança não perde
Do Neto da clara espuma:
Capitão de graça summa
André Cavallo sahiu:
Logo o povo se sentiu;
Porque de incidente novo
Os olhos levou do povo
Quando no Terreiro o viu.
Num branco bruto corria
Mais ligeiro do que o vento,
Tanto que c’o pensamento
Correr parelhas podia:
Veloz desapparecia
Das pernas ao leve abalo,
E não podia julga-lo
O povo que alli se achava,
Si era vento que levava
Pelos ares o Cavallo.

Poz André com bizarria

Todas as lanças mui bem,
E inda assim não faltou quem
Murmurasse todavia.
Soube elle da zombaria,
Que se fez e presentiu
Quem fôra o que alli se riu,
E no outro dia com brio
Um cartel de desafio
Fixou, mas ninguem sahiu.

No cartel que poz, mostrava

Que a qualquer que se julgassem
Tres lanças que se tirassem,
Mil cruzados offertava:
O delinquente acceitava
O desafio esta vez,
Porém que sem interêz
Com gosto perder queria
Nesta contenda e porfia
Não só mil cruzados, tres.
Pede licença ao senhor,
Que no nome a graça traz;
Mas elle como sagaz
O aconselha com primor:
Diz-lhe que fôra melhor
Esta contenda escuzar;
Porém o mancebo alvar,
Fiado em ser cavalleiro
E fiado em ter dinheiro,
Não quiz o pacto acceitar.

Porque se não vence não;

Dizia o moço magnata,
Nem por ouro, nem por prata
O seu sangue de Aragão;
E vendo o senhor dom João
Que si a licença negava
A André Cavallo ultrajava,
Pois podiam presumir,
Si ao Campo o não vissem ir,
Que o dinheiro lhe faltava.

Lhe disse que não só tres

Si corressem mil cruzados,
Si não que depositados
Tinha André Cavallo dez:
Mas o moço Aragonez
Vendo esta resolução,
Por temer a perdição,
A que punha o seu dinheiro,
Toma conselho primeiro
Co’o reverendo Frizão.
O padre, que sem estudo
As leis entende civis,
E com manhosos ardis
Obra mal e entende tudo;
Lhe diria mui sizudo
Com aspecto venerando,
Rindo-se de quando em quando,
Que assim seus enganos lavra:
«Não se lhe dê da palavra,
Diga que estava zombando.»

Assim foi que o desafio

Veiu a parar em burrada,
Que a palavra não val nada,
Si na occasião falta o brio:
E para que com desvio
Não fossem mais inimigos,
Evitando alguns perigos,
Em boa paz os chamou
O General e tractou
De que ficassem amigos.

Depois das pazes, emfim

Lhes pediu que cavalgassem,
E um par de lanças tirassem
Cada qual em seu rossim:
Elle lhe disse que sim,
E de improviso avisou
Ao irmão que não tardou
Com trazer-lhe bons arreios,
Cavallos, sellas e freios,
E com elles se embarcou.
Num dia dos derradeiros
Ao Terreiro os dous chegaram,
E ambos se separaram
Logo dos mais cavalleiros:
Cuidam que são os primeiros
Fidalgos que a terra tem;
E nescios não antevem
Que diz o povo, e não erra,
Que são Fidalgos da Terra,
E outros na Terra ha tambem.

Empinou-se-lhes a Ruça,
E de quatro companheiros,
Sem mais outros cavalleiros,
Fizeram a escaramuça:
O General se debruça,
Para mette-los bem nella,
Na janella com cautela;
Porém usou de revoltas,
Porque mettendo-os nas voltas,
Mandou fechar a janella.

A escaramuça acabada
Fizeram a cortezia,
E todo o povo seria
Vendo a janella fechada:
Nas voltas não viram nada;
Que com notavel trabalho
No ay hombre cuerdo á cavallo;
Porém depois que acabaram,
E o General não acharam,
Ficaram de vinha d’alho.
Com rostos descoloridos,
Desesperados agora
Iam por dentro e por fóra
Da propria côr dos vestidos:
Os que são desvanecidos
E, sem prudencia e razão,
Presumem mais do que são,
Emendem seus pensamentos,
Que para seus desalentos
É vivo o senhor d. João.

Não presumam porque tem,

Que são mais que os pobres, nobres,
Pois ha muitos homens pobres
Mui bem nascidos tambem:
Ao pequeno não convém
Por pequeno desprezar,
Que si este quizer fallar,
Achar póde algum defeito,
Que nenhum ha tão perfeito
Em quem se não possa achar.

Seguia-se um cavalleiro
Ao famoso André Cavallo,
Que levou sem intervallo
De cada golpe um carneiro:
Tambem foi aventureiro
De um premio, mas com defeito
Dava ao corpo um grande geito,
E ficou passado e absorto
De que fosse ao premio torto
E o premio a outro direito.
Ao famoso Braz Rabello
Razão é de mestre o apode,
Pois dar dias sanctos póde
Nesta arte ao que fôr mais bello;
E si com louco desvelo,
Do que digo algum se abraza,
Attenda á razão que é raza,
E verá se faz espantos
Que dar possa os dias sanctos
Quem tem Domingas de Casa.

Nas lanças que poz mui bem

Teve de premios ganança,
E certo que pela lança
Não o ha de vencer ninguem:
Dos cavalleiros que tem
Modernos hoje a Bahia
Leva Braz a primazia,
Porque não ha nesta Praça
Quem se ponha com mais graça,
Fortaleza e bizarria.

Tambem aquella fatal

Emulação de Mavorte,
Para os inimigos forte,
Para os amigos leal;
Applauso merece egual;
Pois nesta cavallaria
Si aos mestres não excedia
Por mais antigos nesta arte,
Aos modernos nesta parte
Elle leva a primazia.
Tambem no Machado fallo,
Que é razão por elle accuda,
Pois sempre ao Cavallo ajuda,
Mas não o ajuda o Cavallo:
Inda assim posso louva-lo
Dando-lhe varios apodos,
Porque conheço em seus modos,
E mui bem posso affirmar
Que nisto de cavalgar
Leva vantagens á todos.

Em mau cavallo corria,

Mas um premio mereceu:
Veja-se quem o perdeu
Que cavalleiro seria?
Aposto que alguem diria,
Vendo que ás carreiras passa
Sem fortaleza, nem graça,
Que o moço com seu sendeiro
É nos fumos cavalleiro,
Porém não cá para a Praça.

Outro cavalleiro airoso

Andou na festividade,
E vi na velocidade
Com que corre ser Velloso;
Por cavalleiro famoso
O povo o acclamou de novo,
E eu só admirando o louvo,
E acho discrição calar,
Porque é escusado fallar
Quando por mim falla o povo.
O Ricardo valeroso
Andou bem, porém sem sorte,
Porque tem pouco de forte,
Si bem tem muito de airoso:
Perdeu pouco venturoso,
Mas sem nenhum sentimento,
Um premio que Braz attento
Ganhou; porque não se atreva
Á aquillo que tambem leva
Com as palavras o vento.
 CHEGANDO O MARQUEZ DAS MINAS
A GOVERNAR O ESTADO COM O CONDE DO PRADO SEU
FILHO, TRACTOU LOGO DE ALLIVAR OS MAGNATES DA
BAHIA, CHAMANDO-OS DO DESTERRO EM QUE PADECIAM,
AMEDRONTADOS DO SEU ANTECESSOR PELA MORTE QUE
OUTROS DERAM AO ALCAIDE MÓR FRANCISCO TELLES, E
POR ACÇÃO DE GRAÇAS LHE FEZ O SECRETARIO DE
ESTADO BERNARDO VIEIRA RAVASCO ESTA DECIMA, QUE O
POETA GLOZOU COM OS PRIMORES COSTUMADOS
METAPHORICAMENTE

DECIMA

De flores e pedras finas

Floresce e enriquece o Estado,
Floresce sim pelo Prado,
E enriquece pelas Minas[7]:
As aves que peregrinas
Aos montes se retiraram,
Nesta manhã já cantaram
Com tão doce melodia,
Que a noite se tornou dia
Porque as penas se acabaram.

[7] Refere-se ao conde do Prado e ao marquez das Minas.

GLOZA

Já da primavera entrou
A alegre serenidade,
Com que toda a tempestade
Do triste inverno acabou:
Já Saturno declinou
Nas operações malignas:
Com influencias benignas
Jupiter predominante
Nos promette anno abundante
De flores e pedras finas.

Si d’estes aspectos taes

Bem se calcula a figura,
Teremos grande fartura,
Não ha de haver fome mais:
Mostras temos e signaes
De um tempo muito abastado:
Porque bem considerado
D’elle tem o proprio effeito,
Já vemos que a seu respeito
Floresce e enriquece o Estado.

Para ser enriquecido

Este Estado e florescente,
Temos a causa patente
No planeta referido:
Não se equivoque o sentido
No effeito aqui declarado,
Porque sendo bem notado,
O Estado (como parece)
Si pelo mais não floresce
Floresce sim pelo Prado.
Pelo Prado flôr á flôr
Se vai a terra esmaltando,
Com que o clima está mostrando
Temperamento melhor:
Do luminar superior
Por taes influencias dignas,
Sendo as pedras o boninas
Da terra unicos primores,
Pois se esmalta pelas flores
E enriquece pelas Minas.

Na terra já se experimentam
Virações tão temperadas,
Que as aves exterminadas
Tornar aos ninhos intentam:
Já não sentem, não lamentam
Tempestuosas ruinas;
Pois com salvas matutinas
Se mostram tão prasenteiras,
Que mais parecem caseiras
As aves que peregrinas.

Sua peregrinação
Influxo foi de Saturno,
Planeta sempre nocturno,
E muito importuno então:
Todas nessa conjuncção
Si os doces ninhos deixaram,
E tanto se receiaram
Do nocivo temporal,
Que escolhendo o menor mal
Aos montes se retiraram.
Porém tanto que sentiram
Haver no tempo mudança
Sem receio e sem tardança
Aos ninhos se reduziram:
Outros ares advertiram,
Outra clemencia notaram,
Com que alegres publicaram
Dos astros os movimentos,
E com festivos accentos
Nesta manhã já cantaram.

Cantaram para mostrar

Com repetidas cadencias
Singulares excellencias
De um planeta singular:
Tal doçura no cantar
Não se ouviu nesta Bahia,
Ouvindo-se na harmonia
Modulações tão suaves,
Que nunca cantaram aves
Com tão doce melodia.

Cada qual com voz sonora

Nos mottetes que cantaram,
Por mil modos explicaram
De todo o Estado a melhora:
Cada instante e cada hora
A musica mais se ouvia,
No Prado resplandecia
Por modo maravilhoso
Um lustre tão luminoso,
Que a noite se tornou dia.
Entre as aves modulantes,
Que este nosso paiz tem,
Todas cantaram o bem
De que são participantes:
Dos males que foram antes
Todas tambem se queixaram;
Assim que todas mostraram
Com alegrias notorias
Que começaram as glorias
Porque as penas se acabaram.
 A UNS CLERIGOS
QUE INDO AO EXAME DO CANTOCHÃO PARA ORDENS
SACRAS NA PRESENÇA DO ARCEBISPO D. JOÃO FRANCO DE
OLIVEIRA, DESAFINARAM PERTURBADOS

Senhor, os padres d’aqui

Por b quadro e por b mol
Cantam bem re mi fa sol,
Cantam mal la sol, fa mi:
A razão que eu nisto ouvi,
E tenho para vos dar,
É que como ao ordenar
Fazem tanto por luzir,
Cantam bem para subir,
Cantam mal para baixar.

Porém como cantariam

Os pobres perante vós?
Tão bem cantariam sós,
Quão mal onde vos ouviam:
Quando o fa bordão erguiam
Cada um parece que berra,
E si um dissona, outro erra,
Mui justo me pareceu,
Que sempre á vista do céu
Fica abatido o que é terra.
Os padres cantaram mal,
Como estava presupposto,
E inda assim vos deram gosto,
Que eu no riso vi o signal:
Foi-se logo cada qual
Direito ás suas pousadas,
Á estudar nas taboadas
Da musica os sete signos,
Não por cantar a Deus hymnos,
Mas por vos dar badaladas.

Vós com voz tão doce e grata

Enleastes meus sentidos,
Que ficaram meus ouvidos
Engastados nessa prata:
Tanto o povo se desata
Ouvindo os vossos esp’ritos,
Que com laudatorios gritos
Dou em fé que uma donzella
Disse, qual outra Marcella,
O Cantico Benedictus.
EPISTOLA AO CONDE DO PRADO
ROMANCE

D’aqui d’esta praia grande

Onde á cidade fugindo,
Conventual das arêas
Entre mariscos habito:
A vós, meu conde do Prado,
A vós, meu principe invicto,
Illustrissimo Mecenas
De um poeta tão indigno,
Enfermo da vossa ausencia,
Quero curar por escripto
Sentimentos, saudades,
Lagrimas, penas, suspiros.
Ausentei-me d’esta Terra,
Porque esse povo maldicto
Me poz em guerra com todos,
E aqui vivo em paz commigo.
Graças a Deus que não vejo
Neste meu doce retiro
Hypocritas embusteiros,
Velhacos intromettidos.
Não me entram nesta palhoça
Visitadores prolixos,
Politicos enfadonhos,
Ceremoniosos vadios.
Visitam-me o lavrador
Sincero, simples e liso,
Que entra co’a bocca fechada,
E sahe co’o queixo cahido.
Dou na varanda um passeio,
Ouço cantar passarinhos
Docemente, ao que entendo,
 Excepto a lettra e tonilho.
Vou-me logo para a Praia,
E vendo os alvos seixinhos,
De quem as ondas murmuram,
Por mui brancos e mui limpos,
Os tomo em minha desgraça
Por exemplo expresso e vivo,
Pois eu por limpo e por branco
Fui na Bahia mofino.
Queimada veja eu a Terra
Onde o torpe idiotismo
Chama aos entendidos nescios,
E aos nescios chama entendidos.
Queimada veja eu a Terra,
Onde em casa e nos corrilhos
Os asnos me chamam asno;
Parece cousa de riso.
Eu sei de um clerigo Zote,
Parente em grau conhecido
D’estes que não sabem musa,
Mau grego e peior latino,
Ambicioso avarento,
Das proprias negras amigo,
Só por levar a gaudere
O que aos outros custa gimbo;
Que si acaso em mim lhe fallam
Torcendo logo o focinho,
«Não me fallem neste asno»,
Responde em todo o seu sizo.
Tambem sei que um certo Beca,
No Pretorio presidindo,
Onde é salvage em cadeira,
Me pôz asno de banquinho.
Por signal que eu respondi
A quem me trouxe este aviso,
Si fôra asno, como eu sou,
Que mal fôra á esse ministro.