Dissertations in Forestry and Natural Sciences
DISSERTATIONS | ULLA GAIN | FRAMING OF COGNITIVELY COMPUTED INSIGHTS | No 462
ULLA GAIN
Framing of cognitively computed insights
Proofs of concept by cognitive services PUBLICATIONS OF
THE UNIVERSITY OF EASTERN FINLAND
Framing of cognitively computed insights
Proofs of concept by cognitive services
Gain Ulla
Framing of cognitively computed insights
Proofs of concept by cognitive services
Publications of the University of Eastern Finland Dissertations in Forestry and Natural Sciences
No 462
University of Eastern Finland Joensuu/Kuopio
2022
PunaMusta oy Joensuu, 2022
Editors: Pertti Pasanen, Nina Hakulinen, Raine Kortet, Matti Tedre and Jukka Tuomela
Sales: University of Eastern Finland Library ISBN: 978-952-61-4461-0 (print)
ISBN: 978-952-61-4462-7 (PDF) ISSNL: 1798-5668
ISSN: 1798-5668 ISSN: 1798-5676 (PDF)
Author’s address: Ulla Gain
University of Eastern Finland School of Computing
P.O. Box 1627
70211 KUOPIO, FINLAND email: gain@uef.fi
Supervisors: Virpi Hotti, Ph.D.
University of Eastern Finland School of Computing
P.O. Box 1627
70211 KUOPIO, FINLAND email: virpi.hotti@uef.fi
Professor Pekka Toivanen, D.Sc. (Tech.) University of Eastern Finland
School of Computing P.O. Box 1627
70211 KUOPIO, FINLAND email: pekka.toivanen@uef.fi
Reviewers: Tuomo Kujala, Ph.D., Associate Professor University of Jyväskylä, Cognitive Science Faculty on Information Technology 40014 JYVÄSKYLÄN YLIOPISTO, FINLAND email: tuomo.kujala@jyu.fi
Annika Wolff, Ph.D., Assistant Professor Lappeenranta-Lahti University of Technology School of Engineering Science
53850 LAPPEENRANTA, FINLAND email: Annika.Wolf@lut.fi
Opponent: Sanna Kumpulainen, Ph.D., Associate Professor Tampere University, Tampere Research Center for Information and Media
Faculty of Information Technology and Communication Sciences
Kalevantie 4, 33100 TAMPERE, FINLAND email: sanna.kumpulainen@tuni.fi
7 Gain, Ulla
Framing of cognitively computed insights. Proofs of concept by cognitive services.
Kuopio: University of Eastern Finland, 2022 Publications of the University of Eastern Finland
Dissertation in Forestry and Natural Sciences 2022; 462 ISBN: 978-952-61-4461-0 (print)
ISSNL: 1798-5668 ISSN: 1798-5668
ISBN: 978-952-61-4462-7 (PDF) ISSN: 1798-5676 (PDF)
ABSTRACT
In business practices, there is a need for new tools to manifest insights from data. There is an ongoing research gap in assessing what benefits new tools (e.g., cognitive services) offer, what they can automate, and how they affect human behaviour. This dissertation aims to fill the gap with answers to the following research questions: What are the correspondences between human cognition and cognitive services?; How to amplify human cognition within the cognitively computed insights?; and How can the impacts of the cognitively computed data for organisations be assessed? Cognitively computed insights are manifested either by cognitive services or automated machine-learning (ML) frameworks.
The main results included 20 constructions: four abstractions, seven experiments, three frameworks, and six mappings. The term "construction"
refers to a typed entity, the types of which are abstraction, experiment, framework, and mapping. The main meanings of the typed constructions are vocabulary based, and their meanings are illustrated in the dissertation as follows: abstraction is relevant information used to highlight a specific purpose; experiment is a purposive investigation to gain experience and pieces of evidence; framework is a set of reusable elements used to guide
8
solution-focused development; and mapping is an assigned correspondence between two entities to explain differences and similarities.
The constructions were established within existing frameworks, which helps reveal interest in the focus of this thesis. The constructions were the result of adapting several frameworks concerning automated machine learning (e.g., Pycaret), brain models (3D Brain, the layered reference model of the brain), business model canvases (e.g., the value proposition canvas), McKinsey’s automation capabilities, personality traits and types (global vectors for word representation, International Personality Item Pool), and value propositions (e.g., Bøe-Lillegraven’s ambidexterity value chains).
Moreover, the following cognitive services exemplify the constructions: IBM Personality Insights, IBM Tone Analyzer, IBM Natural Language Understanding (previously Alchemy Language), IBM Retrieve and Rank, IBM Tradeoff Analytics, IBM Discovery, IBM Visual Recognition, and Microsoft Speaker Recognition.
First, correspondences between human cognition and cognitive services were explored. Eight constructions (one abstraction, three experiments, one framework, and three mappings) derived parts of the answer. There are many human cognitive functions that are reasonable to classify into bigger functional entities. Therefore, mapping human cognitive functions onto groups of cognitive functions was performed before the cognitive services were compared with the human cognitive functions. The abstraction to cognitive functions functional hierarchy concerns the similarities between cognitive services and human cognitive functions presented by the 3D Brain model. One hundred thirty-seven human cognitive functions were studied and compared to cognitive services: the IBM Tone Analyzer functionalities were similar to 65 human cognitive functions; the IBM Visual Recognition functionalities were similar to 27 human cognitive functions; and the Microsoft Speaker Recognition functionalities were similar to 45 human cognitive functions. A framework for the functional hierarchy of cognitive functions was established to attain comparability and correspondences between human cognitive functions and functions of the cognitive services.
Two mappings concerned six cognitive services (IBM Natural Language
9 Understanding, IBM Tone Analyzer, IBM Personality Insights, IBM Retrieve and Rank, IBM Tradeoff Analytics, and IBM Discovery). First, eleven out of McKinsey’s 18 automation capabilities concerning work activities were mapped to cognitive services. Second, four processes out of 52 human cognitive processes presented by the layered reference model of the brain were used to replace three automation capabilities. Finally, four discovered verbs (bind, facilitate, revise, and manifest) were encapsulated through the experiment concerning rules of the capabilities provided using cognitive services to facilitate human cognition. The workflow experiment for value propositions concerned five cognitive services (IBM Alchemy Language, IBM Natural Language Understanding, IBM Tone Analyzer, IBM Personality Insights, and Microsoft Text Analytics). It is possible to derive outcomes from text without human intervention. However, the International Personality Item Pool framework was used in the experiment to generate the rules for transforming personality traits into questionnaires. The main aim was to find the ground truth concerning the value proposition of the cognitive services to support human cognition. Awareness of both capabilities and functionalities of the cognitive services can contribute to fulfilling the system or software system requirements.
Second, the methods for amplifying human cognition within cognitively computed insights were collated. Six constructions (one abstraction, three experiments, and two mappings) derived parts of the answer. In the early stages of the research, big data analytics were used instead of cognitive services. Therefore, the first review concerned big data analytics. The review results formed the basis for the abstraction of uncovering information nuggets from heterogeneous data as part of competitive advantage that was constructed to understand what can be calculated and what is worth calculating, and why. Further, the key questions were introduced for data milling to find indicators. One of the first cognitive services was the IBM Personality Insights, which was used in two mappings: mapping between principles of business analytics and personality insights created transparency between business analytics measurements and refined personality insights; mapping between personality traits and expected
10
experience helped to understand customer experience based on personality traits. The IBM Personality Insights service is an API (application programming interface) service based on one of the most famous word- embedding algorithms: Glove (global vectors for word representation). The experiment included a data dump of 20 web channels (e.g., Facebook comments) containing 53,294 messages to develop transparency and understanding concerning corpus-based insights by the IBM Personality Insights service. However, it was impossible to explain explicitly and exactly how the IBM Personality Insights service calculates the values of the traits.
In the next experiment, semantic roles (subject, action, object) were coded from the General Data Protection Regulation (GDPR) by a human interpreter and the IBM Natural Language Understanding service. Krippendorff’s alpha value was 0.85, which indicates that the capabilities of the IBM Watson Natural Language Understanding service can be used to amplify human interpreters. Cognitive services not only manifest cognitively computed insights. An automated machine learning frameworks (e.g., Pycaret) were reviewed to discover their capabilities, especially within business intelligence (BI) tools (e.g., Microsoft Power BI). An experiment concerning the low-code autoML-augmented data pipeline revealed a lack of interoperable low-code autoML frameworks within BI tools. Only Pycaret was adaptable in Microsoft Power BI. The outcomes of the cognitive services and automated machine learning frameworks can be used as building blocks to construct more significant functional entities. Further, cognitive services can enhance insights on a product, process, or service, and therefore they can be targeted to meet the needs of stakeholders and companies.
Third, ways to assess the impacts of the cognitively computed data for organisations were gathered. Six constructions (two abstractions, one experiment, two frameworks, and one mapping) derived parts of the answer. Three constructions outlined data-driven performance management: the experiment concerning information nuggets for indicators emphasised the importance of key questions; the abstraction of two-way transparency between selected data and principles emphasised
11 performance monitoring; and the mapping between business analytics and ambidexterity value chains emphasised value proposition. Usually, performance management is based on derived data. Both real-time and inferred insights are required. Therefore, measurable advantages in digitalisation are proposed to ensure that the framework of situational information and the possibilities of the inferred insights is facilitated with the abstraction of the utilisation mindset of cognitive service outcomes. Two cognitive services (IBM Visual Recognition and Microsoft Speaker Recognition) were used to build the framework of context-aware information transparency and smart indicators. The framework supports the industry in combining Industrial Internet-of-Things (IIoT) business models and value propositions to match the intelligent insights of cognitive solutions to business objectives.
Once the constructions were conducted to answer the research questions, five research process phases (i.e., hype framing/landing, functional framing, content framing, technical framing, and continuous impact assessment) were established. The research process phases can be adapted in proofs of concept. They can be used in organisations to establish proofs of concept before adapting new building blocks or new technology.
From the framing can be observed how the need for human cognition amplifications affects how the impacts can be assessed. In conclusion and based on the results, the impact of well augmented, transparent, objective, and actionable cognitive data usage with or without interventions is built in these arguments’ nature, result in usage impact is self-evident. However, the organisations must make their experiments to identify their competitiveness and effectiveness by adopting objective insights with the help of cognitive services. The outcomes of the cognitive services can be used as building blocks to construct more significant functional entities.
Furthermore, the research can propose constructions to assess the impacts of the cognitively computed insights.
Keywords: Cognitive computing, cognitively computed insights, proofs of concept by cognitive services, cognition, assessment
12
13
ACKNOWLEDGEMENT
Based on the goal of cleaning up, processing, and digging into data to find the relevant information nuggets, the hunger for insight has grown with the development of technology to obtain something cognitively useful from data. This journey has been long and absorbed into one researcher’s life and has resulted in the curiosity to search for explanations and seek answers.
Now it is time to thank those who helped. I would like to thank Virpi Hotti, PhD, and Pekka Toivanen, D.Sc. (Tech.), for their expert advice and encouragement throughout this challenging project. I want to thank the reviewers, Docent Tuomo Kujala and Assistant Professor Annika Wolf, for their constructive advice that has helped improve this thesis. Thank you, Assistant Professor Sanna Kumpulainen, for being my opponent. At the same time, I would like to thank my family and friends for their encouragement that came when I needed it; thank you for believing in me.
Kuopio, February 18th, 2022 Ulla Gain
14
15
LIST OF ORIGINAL PUBLICATIONS
This thesis is based on data presented in the following articles, referred to by the Roman Numerals I–IX.
I. Hotti V, Gain U. (2013). Big Data Analytics for Professionals, Data- milling for Laypeople. World Journal of Computer Application and Technology, 1(2):51-57, Horizon Research Publishing.
II. Hotti V, Gain U. (2016). Exploitation and exploration underpin
business and insights underpin business analytics. Communications in Computer and Information Science, 636:223-237, Springer, Cham.
III. Gain U, Hotti V, Lauronen H. (2017). Automation capabilities challenge work activities cognitively. Futura, 36(2):25-35.
IV. Gain U, Hotti V. (2017). Tones and traits - experiments of text-based extractions with cognitive services. Finnish Journal of EHealth and EWelfare, 9(2-3):82-94.
V. Gain U. (2020). The cognitive function and the framework of the functional hierarchy. Applied Computing and Informatics, Emerald Publishing Limited, 16(1/2):81-116. DOI:
https://doi.org/10.1016/j.aci.2018.03.003
VI. Gain U, Koponen M., Hotti V. (2018). Behavioral interventions from trait insights. Communications in Computer and Information Science, 907:14-27, Springer, Cham.
16
VII. Gain U, Hotti V. (2020). Awareness of automation, data origins, and processing stakeholders by parsing the General Data Protection Regulation sanction-based articles. Electronic government. DOI:
10.1504/EG.2021.10034597
VIII. Gain U, Hotti V. (2021). Low-code autoML-augmented Data Pipeline – A Review and Experiments. Journal of Physics: Conference Series 1828012015.
IX. Gain U. (2021). Applying Frameworks for Cognitive Services in IIoT.
Journal of Systems Science and Systems Engineering 30: 59-84, Springer Nature, DOI: 10.1007/s11518-021-5480-x
17
AUTHOR’S CONTRIBUTION
The idea of framing the cognitively computed insights is the author's.
Framing combines the constructions and findings of the research papers (I–
IX). The contributions of the authors are described as paper based.
I. The research idea was to improve understanding of big data and the possibilities of data analysis. Gain and Hotti participated equally in the literature appraisal, analysis, and writing process and formed the proposed constructions.
II. The research idea was based on the shared knowledge of the authors (Gain and Hotti) that behaviour management is an important management area in business. Therefore, the main aim of Paper II was to encourage experiments around the behaviour-centric value proposition based on objective insights. The possibilities (e.g., ambidexterity and consciousness) of cognitively computed insights, especially using cognitive services, are illustrated by the recently published IBM Personality Insights service that offers personality traits and consumption preferences. Gain and Hotti participated equally in the literature appraisal, analysis, and writing process and formed the proposed constructions.
III. The research idea was based on the shared knowledge of the authors Gain, Hotti, and Lauronen. Both McKinsey's automation capabilities and Wang’s cognitive processes offer definitions that can be used to classify cognitive services. The corresponding author (Gain) was the main contributor to the utilisation mindset and formalising rules based on the evaluated outcomes of the cognitive services. Hotti and Lauronen participated in the literature appraisal, analysis, and writing process.
IV. The research goal was to determine whether there is ground truth behind the cognitive services, especially IBM Personality Insights. This is based on the shared doubts of authors Gain and Hotti. The authors
18
doubted whether there is non-repudiated argumentation behind the cognitive service extractions (e.g., tones and traits). Gain and Hotti participated equally in the literature appraisal, analysis, and writing process. They formed the proposed constructions, for example using the semantic roles to exemplify the trait-based personality questionnaire.
V. The author was the sole contributor to this publication. The research goal was a deeper understanding of cognitive functions, which can therefore help to better understanding cognitive services. The author performed all the data analysis, interpretation of the results, and writing.
VI. The research idea concerned the techniques behind cognitive services, especially the IBM Personality Insights service, based on experiments involving the cognitive service application programming interface (API) adaptions by all authors (Gain, Koponen, and Hotti) of the paper. There were and still are doubts as to whether text-based trait insights are reliable. The corresponding author (Gain) was a major contributor to the study of word-embedding techniques. The second author (Koponen) implemented the Python program to obtain the IBM Personality Insights service's API-based results and participated in practical data analytics, graphical elements, and the correlation matrix. All authors co-wrote the final manuscript.
VII. This research regarding automation, data origins, and processing in the general data protection regulation sanction-based articles is essential for compliance and needs to be part of proofs of concept in cognitive services. The idea of the research is to supplement the assessment aspect concerning cognitively computed insights. The corresponding author (Gain) was a major contributor to the study of indicative semantic roles. Both authors Gain and Hotti analysed the IBM Watson Natural Language Understanding Text Analysis cognitive service results. Both authors co-wrote the final manuscript.
VIII. The research idea was to review and experiment with AutoML frameworks and provide insight-driven data pipelines where data is
19 ingested, unified, mastered, and enriched as a ground for reports, dashboards, and applications based on the shared knowledge of authors Gain and Hotti.
IX. The idea of applying frameworks for cognitive services in IIoT was the author’s, and she was the only contributor to this publication. The author performed all the data analysis, interpretation of the results, and writing.
These publications are referred to as Papers I–IX throughout this thesis. The above publications have been included at the end of this thesis with their copyright holders’ permission. The permissions to re-publish also on each cover sheet of the published papers.
20
21
TABLE OF CONTENTS
ABSTRACT ... 7 ACKNOWLEDGEMENT ... 13 LIST OF ORIGINAL PUBLICATIONS ... 15 AUTHOR’S CONTRIBUTION ... 17 TABLE OF CONTENTS ... 21 1 INTRODUCTION ... 27 1.1 Research gap and research questions ...28 1.2 Types of constructions...30 1.3 Structure of the dissertation ...32 2 CONCEPTUAL CONTEXT ... 33 2.1 Ambidexterity ...33 2.2 Cognitive computing ...36 2.3 Insights ...45 2.4 Cognitive services ...48 3 SUMMARY OF PAPER-BASED CONSTRUCTIONS ... 57 3.1 Big data analytics for professionals, data milling for laypeople (Paper I)57 3.2 Exploitation and exploration underpin business, and insights underpin
business analytics (Paper II) ...58 3.3 Automation capabilities cognitively challenge work activities (Paper III) .60 3.4 Tones and traits: experiment of text-based extractions with cognitive
services (Paper IV) ...64 3.5 The cognitive function and framework of the functional hierarchy (Paper V)
...65 3.6 Behavioural interventions from trait insights (Paper VI) ...70 3.7 Awareness of automation, data origins, and processing stakeholders
through parsing the general data protection regulation sanction-based articles (Paper VII) ...71 3.8 Low-code autoML-augmented data pipeline: a review and experiments
(Paper VIII) ...72 3.9 Applying frameworks for cognitive services in IIoT (Paper IX) ...73
22
3.10 Research methods as adapted frameworks ...73 4 CONCLUDING REMARKS ... 77 4.1 Power of the research process phases ...81 4.2 Answers to the research questions ...84 4.3 Practical implications ...89 4.4 Future research issues ...92 5 BIBLIOGRAPHY ... 95 6 PAPERS ... 105
23 LIST OF TABLES
Table 1. Mapping between business analytics principles and personality insights (adapted from Table 2, Paper II: table content errors have been fixed). ... 59 Table 2. Cognitive services versus capabilities of cognitive services. 62 Table 3. Constructions and used cognitive services: PI = IBM Personality
Insights, TA = IBM Tone Analyzer, NLU = IBM Natural Language Understanding, AL = Alchemy Language, RR = IBM Retrieve and Rank, TAO = IBM Tradeoff Analytics, D = IBM Discovery, VR = IBM Visual Recognition, and SR = Microsoft Speaker Recognition.77 Table 4. Context-related constructions: A = ambidexterity, CC= cognitive
computing, I = insights. ... 78 Table 5. Constructions and the main aim of utilisation as well as
research questions: P = paper, A = abstraction, E = experiment, F = framework, M = mapping, Q1= cognitive capabilities, Q2=
human cognitive amplification, Q3= impact assessment. .... 84
24
25 LIST OF FIGURES
Figure 1. The proofs-of-concept cycles from data to insight. ... 29 Figure 2. The transformation from data to information, knowledge, and
wisdom through cognition. ... 37 Figure 3. The architecture of cognitive computing, adapted from Kaufman
et al. (2015). ... 43 Figure 4. The example of Peirce’s sign (adapted from Hiltunen, 2010). I
came out from the cottage to the terrace and looked left into the woods. Then, my attention (i.e., vision) attaches to the view presented in the triangle (I became frightened). ... 46 Figure 5. The stump in the morning light. I calmed down as it did not
seem to be moving. I decided to go closer and found a stump.
... 47 Figure 6. Outcomes of the cognitive services. ... 54 Figure 7. IBM Personality Insights: big-five personality dimensions and 30
facets (adapted from IBM, 2021a). ... 54 Figure 8. IBM Personality Insights: 12 needs and five values (adapted
from IBM, 2021a). ... 55 Figure 9. IBM Personality Insights: 42 consumption preferences (adapted
from IBM, 2021a). ... 55 Figure 10. Utilisation mindset of cognitive computing (Paper III, Figure 4).
... 64 Figure 11. Framework for the functional hierarchy of cognitive functions (adapted from Paper V). ... 66 Figure 12. Hierarchy of cognitive functions (adapted from Paper V, Figure
4). ... 68 Figure 13. Research process phases and research questions framed for
construction mappings... 82
26
LIST OF ABBREVIATIONS
3D Three dimensional AI Artificial intelligence API Application interface
AutoML Automated machine learning BI Business intelligence
Glove Global vectors for word representation IE Information extraction
IoT Internet of Things
IPIP International Personality Item Pool
DIKW Data, information, knowledge, wisdom model LDA Latent Dirichlet allocation
NED Named entity disambiguation NEL Named entity linking
NEN Named entity normalisation NER Named entity recognition
NERD Named entity recognition and disambiguation NLP Natural language processing
NDC Not defined cognitive
PLSI Probabilistic latent semantic indexing Q/A Question answering
TF-IDF Term frequency inverse document frequency
UIMA Unstructured information management architecture WDA Watson discovery advisor
27
1 INTRODUCTION
The evaluation of data and information is referred to as business intelligence (BI), and it partly aims to judge performance. In the era of digitalisation, data discovery-oriented platforms are the mainstream in BI. Thus, there is an urgent need for a new generation of computational theories and tools to assist business users in extracting useful information and insights from structured and unstructured data. Understanding what tools may support different parts of a data pipeline is essential for creating insights from data and identifying common drawbacks of the employed tools: they often lack adequate support for laypeople (i.e., non-experts in data systems) such as selecting the correct parameters for setting up the tool. This disadvantage relates to several literacy-related fields including data literacy, machine- learning (ML) data science literacy, and general computing literacy.
One way to bring the insights closer to the experiments in business is to use the same concepts. Further, ideas around organisational ambidexterity (i.e., exploitation and exploration) have been adapted within automation when organisations confront the problems of inadequate resources.
Cognitive services are integrated into smart things, and they bring knowledge and a learning environment to BI in order to increase human cognition. Smart things refer to things that participate on the internet of everything (Langley et al., 2021). Those are constructed by different combinations of signals, hard and software smart functionality that can connect people, systems, processes, and assets, including monitoring, control, optimisation, and autonomy. Notably, cognitive services can be embedded in more extensive functionality and, in practice, produce their functions in a discreet manner so that the end user is not even aware of using the core capabilities of cognitive computing (Kelly, 2015b). For example, the Talkspace online therapy service was constructed by adapting the IBM Personality Insights Service as it manifests personality traits from
28
textual data (Talkspace, 2020; IBM, 2021a). Grammarly is a browser extension that manifests tones from textual data (Grammarly, 2020), research process phases, and research questions.
1.1 Research gap and research questions
In business practices, there is a need for new tools to gain insights from data.
However, there is an ongoing research gap in assessing what benefits the new tools (e.g., cognitive services) offer, what they can automate, and how they affect human behaviour. Deployments of new technologies are being delayed due to a lack of understanding of process phases, from hype1 or unknown items to their impact assessments.
Figure 1 presents the fundamental entities of this thesis (i.e., data, technology, and human cognition). It illustrates proofs-of-concept cycles for discovering the insights gained in this thesis. Correspondences between human cognition and cognitive services (i.e., transparency) are strongly related to the amplification of human cognition and its assessment since we need to amplify or reject without correspondences. In other words, human cognition needs to be understood to build objective correspondences with data. Arguments need to be presented that construct transparency which can be verified. Cognitively computed insights are manifested either by cognitive services or automated ML frameworks.
1 hype refers a situation in which something is brought out to attract everyone's interest (Cambridge University Press, 2021)
29 Figure 1. The proofs-of-concept cycles from data to insight.
Organisations must be exploratory and exploitative, which means that continuously proofs of concept must be addressed for new ideas or hype issues. Some ideas or issues are slowly ripening. Therefore, deployments of new technologies are delayed because there is no natural adaption framework for organisations, and their impact is not addressed by proofs of concept. This dissertation aims to fill the gap with answers to the following research questions: What are the correspondences between human cognition and cognitive services?; How to amplify human cognition within the cognitively computed insights?; and How can the impacts of the cognitively computed data for organisations be assessed?
30
1.2 Types of constructions
The International Organization for Standardization's Online Browsing Platform (ISO OBP), the Software and Systems Engineering Vocabulary (SEVOCAB), and the Unified Compliance Framework Dictionary (UCF) are three vocabularies that were used to obtain evidence for definitions. The vocabularies were selected based on meaningful content, that is, they cover terms used in software system domains and authority documents (e.g., laws and standards). The IEEE Computer Society and ISO/IEC JTC 1/SC7 constitute the SEVOCAB authors, and they collect terms concerning software system domains. Partly, SEVOCAB contains terms of the ISO Online Browsing Platform (ISO OBP), mainly a collection of standardised terms. The UFC collects terms concerning authority documents2, and the terms are usually general without domain specificity.
The term "construction" is associated with implementation (SEVOCAB), a
"process of writing, assembling, or generating assets" (ISO OBP, SEVOCAB), a structure or "complex entity" that is "made of many parts" (UCF). Further, an implementation requirement is defined as a “construction of a system or system component” (SEVOCAB), and a construction element is a "constituent of a construction entity with a characteristic function, form, or position" (ISO OBP).
The term "construction" is meant to be a typed entity, the types of which are abstraction, experiment, framework, and mapping. The term
"abstraction" is defined as a "[p]reoccupation with something to the exclusion of all else" (UCF) and a "view of an object that focuses on the information relevant to a particular purpose and ignores the remainder of the information" (ISO OBP, SEVOCAB). The term "experiment" refers to trying "something new, as in order to gain experience" (UCF), or it is a
"purposive investigation of a system through selective adjustment of
2 Authority documents refer to document types as follows: Statutes, regulations, directives, principles, standards, guidelines, best practices, policies, and
procedures (UCF)
31 controllable conditions and allocation of resources" (ISO OBP). The term
"framework" is defined, for example, as a "real or conceptual structure used as a support or guide for building something" (UCF), a "particular set of beliefs or ideas referred to in order to describe a scenario or solve a problem" (ISO OBP), or a "reusable design (models or code) that can be refined (specialised) and extended to provide some portion of the overall functionality of many applications" (SEVOCAB). The term "mapping" is an
"assigned correspondence between two things represented as a set of ordered pairs" (SEVOCAB), "[a]ny mathematical condition relating each argument (input value) to the corresponding output value" (UCF), or a "set of values having defined correspondence with the quantities or values of another set" (ISO OBP).
The main results of this dissertation are 20 constructions (Chapter 3):
seven experiments, four abstractions, three frameworks, and six mappings.
The main meanings of the typed constructions are vocabulary based. Their meanings as used in this dissertation are as follows: abstraction refers to relevant information used to highlight a specific purpose; experiment is a purposive investigation in order to gain experience and evidence;
framework is a set of reusable elements used to guide solution-focused development; and mapping is an assigned correspondence between two entities in order to explain differences and similarities.
The constructions were established by adapting several frameworks (Section 3.10). When the term “framework” is used to illustrate research materials or methods, its meaning covers content analyses tools and techniques, review guidelines, reliability and validity instructions, as well as contextual ground materials concerning automated ML (e.g., Pycaret), brain models (3D Brain, the layered reference model of the brain), business model canvases (e.g., the value proposition canvas), McKinsey’s automation capabilities, personality traits and types (Global vectors for word representation, International Personality Item Pool), and value propositions (e.g., Bøe-Lillegraven’s ambidexterity value chains).
32
1.3 Structure of the dissertation
This thesis is based on nine scientific papers. Paper I illustrates the examination of information nuggets for better competitive advantage:
possibilities for deeper understanding can lead to reactions. Paper II clarifies the meaning of objective insights such as those cognitively computed. Paper III presents the capabilities of cognitive services. Paper IV clarifies utilisations of the cognitive services with and without interventions. Paper V explicates the correspondences between human cognition and cognitive services.
Paper VI addresses word embeddings concerning textual inputs. Paper VII proposes a method for assessing whether cognitive services are useful in manifesting indicative semantic roles. Paper VIII presents a pipeline from raw data to insights, possibilities of low-code autoML cognitive supportive insights, and deeper understanding of meaningful fields. Paper IX proposes value additions in the form of questions and answers.
The thesis contains four chapters and nine publications, and it is organised as follows. Chapter 1 introduces the research scope and research questions. Chapter 2 presents the conceptual context and cognitive services, the insights of which were researched. Chapter 3 summarises the contributions of the research papers (Papers I–IX) that make up this thesis.
Finally, Chapter 4 concludes the thesis by presenting validity issues, implications, and issues for future research.
33
2 CONCEPTUAL CONTEXT
As the thesis frames cognitively computed insights, the conceptual context of the thesis and research issues are related with ambidexterity (Section 2.1), cognitive computing (Section 2.2), insights in general (Section 2.3), as well as insights computed by cognitive services (Section 2.4).
2.1 Ambidexterity
Ambidexterity refers to organisational capabilities that ensure cash flow and investment in product development. The term “ambidexterity” has its origins in the 1650s and originally referred to the ability to use both hands with equal ease. It was also used in Medieval Latin with the meaning of double dealing (Dictionary.com, 2010). Later, in 1976, Duncan (1976) introduced organisational ambidexterity in business; after that, March (1991) adopted this concept for exploitation and exploration; it illustrates tension in the business model (Raisch and Birkinshaw, 2008). The importance of ambidexterity is highlighted in the context of insight-related issues such as strategic management, innovation, technology management, organisational learning and adaptation, organisation theory, and organisational behaviour (Simsek, 2009).
Ambidexterity of different kinds is used at various levels: structural at the corporate level, contextual at the business-unit level, and sequential at the project level (Chen, 2017). Sequential ambidexterity realigns organisational structure where the focus is shifted temporally between exploitation and exploration to change environmental conditions, strategies, or project-level requirements (Chen, 2017; O’Really and Tushman, 2013). In structural ambidexterity, exploitation and exploration are organised in separate units and coordinated by top managers (O’Reilly and Tushman, 2004; Tushman and O’Reilly, 1996) to use different business-unit strategies, structures, and processes (Chen, 2017). This is compared with contextual ambidexterity, in
34
which the created organisational environment allows employees to freely choose between exploration and exploitation, for example, at Google, engineers can use 20% free time to explore their selected research projects (Chen, 2017).
There is a positive relationship between ambidexterity and firm growth, firm performance, and business-unit performance. Ambidexterity can be explained as a transition of organisational units between two processes—
exploitation and exploration—and by a company’s desire to benefit from the complementarities of the two processes (Zimmermann et al., 2015). As Katila and Ahuja (2002) explained, the exploration of new capabilities (scope) elaborates the knowledge base of the organisation as well as the existing capabilities (depth), which is often needed in exploration.
Organisational duality or tension between exploitation and exploration in organisational learning has been studied through various aspects, for example, Baum et al. (2000) states that “exploitation refers to learning gained via the local search, experiential refinement, and selection and reuse of existing routines. Exploration refers to learning gained through the processes of concerted variation, planned experimentation, and play.”
Incremental innovations are minor adaptions carried out to meet existing customer needs. Radical innovations are fundamental changes made to satisfy emergent customer needs (Raisch and Birkinshaw, 2008).
Incremental innovation can operate in an ambidextrous organisation, and this requires “the fine-grained strategic schemata and the governance principles of the explicit knowledge are the underlying levers that set the dimensions for incremental innovations” (Laukkanen, 2012).
Ambidexterity involves new technologies in strategic management.
Burgelman’s internal ecology strategy model initiatives have an organisational scope and thus increase current knowledge compared to new initiatives that are not part of the scope and therefore require learning (Burgelman, 1988; Raisch and Birkinshaw, 2008). Further, the ambidexterity of ecosystems especially emphasises solutions which search for a balance between exploitation and exploration in the firm ecosystem, for example in Kauppila (2010). The creation of an ambidextrous organisation utilises both
35 exploration and exploitation partnerships. Wan et al.’s (2017) research aimed to "identify the essential tensions regarding platform strategies, and”
and further to “analyse how to balance them within platform ecosystems."
Exploration- and exploitation-based learning, for example planned experimentation and knowledge sharing, are essential to increase organisational capabilities and innovations such as value-proposition-based objective insights. Firms strategically use a wide range of technologies for research and exploitation purposes (Bresciani et al., 2018). Regarding cognitive computing, John Kelly (2015b) propounded that “there is not an industry or discipline that this technology won't completely transform over the next decade.” Furthermore, Lucas and Goh (2009) specified the disruptive technology as follows: “the most important observation is that management has to recognise the threats and opportunities of new information and communications technologies and marshal capabilities for change.” Moreover, for organisational exploitation and exploration, the technical activities need to be linked to the need. In other words, they need to be strategically integrated into the performance management of organisations such as in measures and strategic goals (Simsek et al., 2009;
Burgelman, 1988).
36
2.2 Cognitive computing
Cognition refers to “mental actions or processes of acquiring knowledge and understanding through thought, experience, and the senses” (ISO OBP).
Cognition “enables the new classes of products and services to sense, reason and learn about their users and the world around them” because
“where code and data go, cognition can follow, cognition transforms how a company operates” (Kelly, 2015a). Data are transformed into cognition through the paths of the data, information, knowledge, wisdom (DIKW) model (Wang, 2009; Baškarada et al., 2013). Data are transformed into meaningful information through cognitive processing (Baškarada et al., 2013). Knowledge consists of understood, organised, absorbed, and memorised information; it is accumulated learning. Knowledge consists of skills and experiences acquired by doing something. Wisdom is accumulated knowledge. From a wide perspective, wisdom is a form of cognitive understanding that evolves with age as skills and knowledge accumulate (Takahashi and Overton, 2005). Characteristic of wisdom is an ongoing balancing between knowing and doubting, concordant with the balance theory of wisdom (Takahashi and Overton, 2005). Wang (2009) defines cognitive information as all internal embodiments, for example experience, knowledge, and skills, that is between “data (sensational inputs)”
and “action (behavioural outputs).”
Figure 2 represents the transformation from data to information, knowledge, and wisdom through cognition. In
Figure 2, the smaller arrows represent the ongoing brain processes, and the bigger loop with the arrows represents the ongoing interactive process from external perceptions.
37 Figure 2. The transformation from data to information, knowledge, and wisdom through cognition.
Evolution towards cognitive systems: question answering systems. Question answering (Q/A) systems are composed of information retrieval, natural language processing, information extraction, knowledge representation, and reasoning (Maybury, 2004). Early Q/A systems used databases to answer users’ questions. The questions were mapped as computable database queries; systems were complex and required expert, hands-on direction to maintain and solve the problems. The era of the Worldwide Web has provided a new source and approach for Q/A systems.
In these open-domain question answering systems, an answer must be found and extracted through text retrieval. Pasca (2007) introduced a new model for answer retrieval which embodies the de facto paradigm for Q/A as the following (Webber and Webb, 2010, 630–654): retrieve potentially relevant documents; extract a potential answer; and return a top answer(s).
The era of cognitive computing has brought Q/A systems to a new level; it has reformed the question-answering model and extended each analysis phase. The Watson Jeopardy Q/A system is a predecessor of the Watson
38
Discovery Advisor (WDA). Hence, Watson Jeopardy served as a starting point in the DeepQA project (Beller et al., 2016; Ferrucci et al., 2010). Ferrucci et al. (2010) present WDA system principles as content acquisition, question and topic analysis, question decomposition, hypothesis generation, hypothesis and evidence scoring, synthesis, final confidence merging, and ranking analysis. The content of each phase of the pipelines is provided in Ferrucci et al. (2010).
Evolution towards cognitive systems: transform unstructured data into understandable form. The ability to transform unstructured data into a computable and understandable form was a major step because most business information consists of unstructured data (Blumberg and Atre, 2003). Also, data expansion is an ongoing process; for example, the Internet of Things (IoT) creates more data. A significant portion of unstructured data consists of textual formats such as email bodies, documents, web pages, and social media data. Academic research in information retrieval and computational models such as a vector space, probabilistic retrieval, and Boolean retrieval models laid the groundwork for the progress of search engines (Salton, 1988). Part of the development was the result of computational linguistics techniques such as statistical natural language processing (NLP) for lexical acquisition, word sense disambiguation (WSD), probabilistic context-free grammars, and speech tagging (Manning and Schültze, 1999).
Evolution towards cognitive systems: information extraction. In the era of big data, text-mining techniques have been enriched in several domains, especially in information extraction. Information extraction techniques like sentiment analysis are used to identify the polarity of a text, providing a positive or negative tone. Further, sentiment analysis categorisation helps to scale the sentiment in a text. Sentiment classification and categorisation are problematic causes of subjectivity (Pang and Lee, 2008). Also, detecting sarcasm and irony in a text is challenging (Peng et al., 2015) as is determining the context of a text where a negative or positive sentimental tone is detected. The information extraction (IE) technique called entity recognition (NER), is also known as entity extraction and includes processes to specify
39 names, places, dates, and organisations. It identifies and classifies the entities in the text. There are different ways to exploit NER results; for example, they can be indexed, linked off. They can be used in an IoT of IE relations because “an IoT of information extraction relations is the associations between named entities” (Manning, 2017). They can attribute the sentiments and used them for question answering when the answers are named entities. (Manning, 2017.) The IoT devices and sensors serve as the machine’s senses of the environment; they offer opportunities to leverage value from the unstructured data in the IoT when combining services such as IBM Watson with, for example, the cognitive IoT in fitness and well-being (Sheth, 2016).
Cognitive computing differs from traditional programmed solutions. Among the differences between traditionally programmed solutions, the outcomes of which are pre-determined, and cognitively computed solutions that are probabilistic is their technological capabilities. For example, the system can learn, recognise patterns, and process natural language and images (Kaltenrieder et al., 2015). The outcomes of cognitive computing are probabilistic (Marchevsky et al., 2017). As part of artificial intelligence (AI), cognitive computing composes a set of services that mimic human brain processes (Chen et al., 2016) such as recognising the speaker, transforming speech to text, and abstracting sentiment (for example, positive, negative, or neutral) from a text such as an SMS message (MS, 2017b; MS, 2020a; MS, 2020b). Hoffenberg (2016) presents the difference between cognitive computing and AI, where AI systems offer solutions. The cognitive system provides information about choices (i.e., it helps the human decide).
Cognitive computing aggregates collective and computational intelligence;
furthermore, there is no distinction between disciplines (Kaltenrieder et al., 2015).
Grounds of cognitive computing. Cognitive computing can be seen as an umbrella term for diverse research fields, models, processes, and algorithms such as machine learning (ML), neural networks, semantic and natural language processing, information retrieval, knowledge representation, and reasoning (Kaltenrieder et al., 2015; Gliozzo et al., 2013).
40
Cognitive computing combines extensive academic research in algorithms and device architecture such as deep learning and platforms inspired by natural neural networks (Nahmias et al., 2013). These services are run on cloud platforms that combine neural networks and sophisticated algorithms such as ML, deep learning, and natural language processing to reach solutions with and without human intervention (ElBedwehy et al., 2014;
Williamson, 2017). Cognitive solution development is an iterative process that concerns model development, analysis, and testing, where ground truth data correspond to model accuracy (Kaufman et al., 2015). The general foundation of cognitive computing consists of the model, hypothesis generation, and continuous learning. Cognitive systems’ typical architecture is presented in Figure 3 as follows (Kaufman et al., 2015):
• Presentation and visualisation. Presentation and visualisation services present data that support a hypothesis. They also support a request when the system requires more information to improve the confidence of the hypotheses. For this purpose, the three main types of services are narrative solutions, visualisation services (graphics, images, gestures, or animation), and reporting services such as structured outputs.
• Corpora and other data sources. The knowledge base for a cognitive system is the corpus (plural corpora), which needs to be defined for building a machine-readable model of a specific domain. This system's knowledge base is used for linguistic analysis, discovering relationships or patterns, answering questions, and delivering insights. Therefore, corpora data sources play an important role in the system implementation, and they can be updated continuously or periodical.
Other data sources the system could need are ontologies, taxonomies, catalogues, structured databases of the specific subject, and acquired information such as images, videos, sensor data, and language-related data voice and text.
• Processing services. Processing services transform external data sources’ language text, video images, audio files, and sensor data into a machine-learnable format.
41
• Analytic Services. Analytics collect techniques for presenting characteristics or relationships in the dataset, for example regression analysis. Standard analytic components can be deployed such as descriptive, predictive, and prescriptive tasks performed by statistical software packages. Advanced analytics includes statistics, data mining, and ML.
• Feature extraction and deep learning. Feature extraction and deep learning are used to collect techniques needed to transform data into a form that captures essential properties.
• Natural language processing. Natural language processing (NLP) is used to process unstructured text. This group of techniques aims to extract meaning from text. The techniques include language identification, tokenisation, lexical analysis, syntax and syntactic analysis, resolving of structural ambiguity, disambiguation of word sense, and semantics.
• Data access, acquisition, metadata, and management services. For better answers and recommendations for decisions, the data sources of the cognitive systems may need to be updated to correspond to the latest domain-specific information. When supplemental information needs to be added to the data sources, it must be identified, acquired, and further transformed to support ML. Data access performs the required analysis by identifying the relevant data.
• Internal data sources. Internal data sources are the structured and unstructured data of organisations.
• Infrastructure/deployment modalities. Infrastructure and deployment modalities consist of the networking, hardware, and storage base for cognitive applications. The major considerations for infrastructure approaches are distributed data management and parallelism.
Distributed data management consists of workload and external data resources. Management needs to provide scalability and flexibility for managing large quantities of data. Parallelism can contribute to the cycle of hypothesis generation and scoring to process multiple hypotheses and scorings simultaneously.
42
• Hypothesis generation and scoring. The hypothesis requires evidence or a knowledge-based explanation of a causal relationship. The cognitive system searches for evidence such as experiences and data as well as relationships between data elements that can support or refute a hypothesis. The system can create multiple hypotheses that underpin the data in the corpus. For example, the hypothesis can be constructed based on a user question when the corpus has trained question/answer pairs. The hypothesis is a construct based on given data, where system search patterns are based on assumptions defined in the system. The hypothesis scoring assigns a confidence level for the hypothesis.
• Machine learning. ML algorithms look for patterns. Pattern similarity comparison of elements such as structure, values, or proximity (closeness) of data can interpret the pattern compared to known patterns. For basic learning, the approach can be used for pattern detection. The choice depends on available data and the nature of the problem to be solved. Therefore, the choice of ML algorithms for the cognitive application follow the same principle. For example, supervised learning is a good candidate if the source data and associations between data elements for the problem exist and if the data patterns can be identified from the data, allowing them to be further exploited. The other approach examples are reinforced learning (system takes action and learns by trial and error) and unsupervised learning (detects patterns, knowledge about the relationships and associations of the patterns).
43 Figure 3.The architecture of cognitive computing, adapted from Kaufman et al. (2015).
Cognitive computing application interfaces. The architecture of cognitive systems provides cognitive service application interfaces (APIs) that can be used as building blocks for the solutions. Cognitive computing companies provide these services as APIs at the developer cloud base. For example, Microsoft gathers these service features into five groups: vision, speech, language, decision (MS, 2021), and one group with AI/ML systems such as IBM Watson Text-to-Speech, IBM Watson Speech-to-Text, IBM Watson Discovery, IBM Watson Knowledge Studio, and IBM Watson Natural Language Understanding (IBM, 2021c). These services are iterative and interactive (PAT Research, 2021).
Measurable responses help determine the acceptable level of response. An artificial neural network does not provide user interpretable reasons for output (i.e., the production rules; Louridas and Ebert, 2016), for example object recognition in which the system infers the rules for identification during the training phase (i.e., states are embedded in the neural network)
44
and through object recognition provides the user with the result (i.e., classified the images; Abadi et al., 2015).
Information extraction and use of models. Useful extractible information from text-based data is performed using models of different kinds. The big five model captures the personality traits that underpin theories of personality (De Raad, 2000). However, the model lacks a commonly accepted solution as it misses the best way to form concepts of and measurements for each domain of the big five. The ability to reply to the big five in different cultures and languages and the optimal way to explore subfactors of the big five is also missing. This raises doubts as to whether significant factors beyond the big five exist and whether the big five should be merged into super factors (Johnson, 2017). The text models are, for example, the solution for categorising the word, sentence, or document and the categorisation problem (in other words, classifying new documents, sentences, and words).
The model uses different incorporations of the techniques and knowledge areas such as NLP, ML, and statistics. Word-embedding models aim to find semantically similar words to capture syntactic and semantic regularities from the text. For example, the Google open-source tool and Word2Vec use topic vectors in neural networks or matrix factorisation to learn the vector representation of the word to predict the other words in a sentence. This vector representation captures the structure of the text and is used for text classification (Merret, 2015). Term frequency-inverse document frequency (TF-IDF) calculates the frequency of a term in a document and its importance relative to the corpus (Rajaraman and Ullman, 2011). Topic modelling uses algorithms to find the main themes in large arrays of unstructured collections of documents. Probabilistic latent semantic indexing (PLSI) is a statistical technique that uses probability to model and co-occurrence data (Hofman, 2013). The ML technique latent Dirichlet allocation (LDA) is a developed version of PLSI (Blei, 2012); it identifies clusters of entities that are related (Earley, 2015).
45
2.3 Insights
A standard definition of insight is a profound and unique knowledge about an entity where knowledge is an “outcome of the assimilation of information through learning,” and an entity is “anything perceivable or conceivable” (ISO OBP). However, the term “insight” is not used in statements of authority documents such as directives and regulations (UCF). An insight is connected with something. In other words, insight is “an instance of apprehending the true nature of a thing, through” accurate, deep, clear, and intuitive understanding (Dictionary.com, 2021; Cambridge University Press, 2021;
Lexico.com, 2018). Synonyms for insight such as perception, apprehension, intuition, and understanding (Dictionary.com, 2021) are based on a subjective perspective. From the perspective of an individual, insight is described as “an understanding about the motivational forces behind one's actions, thoughts, or behaviour; self-knowledge” (Dictionary.com, 2021) or
“especially an understanding about the motives and reasons behind one's actions” (Dictionary.com, 2021; Random House Value Publishing, 1996). In particular, the meaning of insight interconnects terms such as motives, behaviour, self-knowledge, and action.
When we look at the workflow from a sign to an insight (in other words, how the thing matures as an insight), at first we receive (i.e., sense) a signal (i.e., a sign). The signal does not have absolute value unless it is attached to a meaningful context (De Saussure, 1983; Hiltunen, 2010; Baškarada and Koronios, 2013). In semiotics, the sign is divided into an object, representamen, and interpretant (Peirce, 1868; Hiltunen, 2010), where Peirce's sign (i.e., object) is objective, and the interpretation (i.e., interpretant) is subjective since it depends on the receiver of the sign. The representamen is objective and subjective, for example the words for objects in different languages (Hiltunen, 2010).
46
Figure 4.The example of Peirce’s sign (adapted from Hiltunen, 2010). I came out from the cottage to the terrace and looked left into the woods.
Then, my attention (i.e., vision) attaches to the view presented in the triangle (I became frightened).
An individual’s abilities influence interpretation. “The interpretant is an individual’s comprehension of and reaction to, the sign-object association” (Baškarada and Koronios, 2013). Interpretation of the sign can be influenced by cognitive bias and self-knowledge. In this example, the bear’s characteristics, the observation, and regional information, for example the number of bears, are interconnected with the sign in the observer's mind. The reception of a signal is influenced by the individuals' mental filter and experience (Ansoff, 1984; Hiltunen 2010). Also, signal quality can affect interpretation, especially if the signal quality is bad, for example when the signal light of the beacon does not appear sufficiently far away in the fog. In Figure 4, the observation is based on visual recognition in the declining evening sun, which highlights colours and shadows differently (Figure 5).
47 Figure 5. The stump in the morning light. I calmed down as it did not seem to be moving. I decided to go closer and found a stump.
Data is defined in semiotics as a “symbol or a set of symbols used to represent something” (Baškarada and Koronios, 2013). Also, Stephen Few (2015) states that data is a collection of facts; when a fact is true and useful (i.e., it must inform, matter, and deserve a response), it is a signal; otherwise, it is noise. Indeed, Peirce’s representamen of an object, in other words a thing, is anything that takes our attention, for example the image, a pattern in processed heterogeneous data such as one that repeats in processed big data or missing value. In other words, how signals mature into an insight is that at first, a signal (i.e., a sign) is received (i.e., sensed), any kind that takes our attention, for example an image or pattern of the heterogeneous data or missing value. Perception through the senses allows us to receive and process data. Perception consists of “how something is regarded, understood, or interpreted” as well as “intuitive understanding and insight”
(Lexico.com, 2018).
Insight adds value through learning. The prerequisite for learning in the digital era consists of creating useful information patterns and combining data sources. Therefore, the capability to recognise and adapt to convert models has become a key learning task. Data is needed, but it does not need to be known beforehand; instead, finding sources that fulfil the requirements is essential. Artificial intelligence is used with data to “extract
48
meaning, determine better results based on continuous learning, and enable real‐time decision making” (Iafrate, 2018).
Before the term “big data” was introduced, companies used basic analytics (mainly by manually examining spreadsheets) to find insights and trends (SAS, 2021). Organisational insight can be described with the help of the metamodel of TOGAF (the open group architecture framework) (The Open Group, 2021). Organisational insights are connected to motivation.
Drivers that expose factors (i.e., opportunities or constraints) motivate the organisation unit to create goals; this is a driver (in other words, insight) that addresses a traceable goal. An objective is a time-bounded benchmark (i.e., near and midterm measurement points) that defines progress towards a goal. Besides, the measures set performance criteria for the objective.
Consequently, the course of action realises the goal; it is influenced by business capability, and it influences the value stream (e.g., value propositions).
2.4 Cognitive services
Cognitive service are capabilities for manifesting insights (i.e., meaningful information) and event trigger actions based on sophisticated algorithms and ML techniques. Cognitive services have become more prevalent in everyday life. They can be embedded in more extensive functionalities and, in practice, discreetly produce their functions so that the end user is not even aware of using the core capabilities of cognitive computing (Kelly, 2015a). In this context, this thesis presents cognitive services that use unstructured data, for example text input. These are focal in this study because their usability has been assessed in the light of various research foci, for example insightfulness. The outcomes (Figure 6) and short descriptions of these cognitive services are presented in the following sections.
IBM Personality Insights. Personality Insights was deprecated and will be discontinued as of the 1 December 2021. Natural Language Understanding replaces Personality Insights as part of the analytical workflow (IBM, 2021a).
49 The service uses an open-vocabulary approach to infer personality traits from textual input of at least 100 words (The accuracy of the analysis can be improved by providing more words) using three models: big five, needs, and values. The service deploys the word-embedding technique GloVe (global vectors for word representation) to transform the words of the input text into a vector representation. Next, the ML algorithm calculates percentiles and raw scores of the personality characteristics from word-vector representations and consumptions preference scores from the personality characteristics; it generates the scores for each personality trait and optionally the consumption preferences. Normalised scores are computed by “comparing the raw score for the author’s text with results from a sample population” (IBM, 2021a). The service computes personality characteristics as follows: five personality dimensions (openness, conscientiousness, extraversion, agreeableness, and emotional range) and 30 facets (Figure 7), 12 needs and five values (Figure 8), as well as the consumption preferences (Figure 9). Big five describes “how a person generally engages with the word”; needs describe what a person “hope[s] to fulfil when [she] consider[s]
a product or service”; and values (or beliefs) “convey what is the most important to [a person]” (IBM, 2021a). Explanations for both the high and low values are provided for dimensions and facets of the personality (IBM, 2021a). However, explanations are only provided for the high scores of the needs (IBM, 2021a) and values (IBM, 2021a). The service evaluates 12 needs at a high level, and these needs describe the aspects of the product that likely resonate with the author of the inputted text (IBM, 2021a). Values are described as motivating factors that influence the author’s decision making (IBM, 2021a). Consumption preferences are grouped in eight categories. IBM describes each preference briefly (IBM, 2021a). The calculated score of consumption preferences indicates the author’s likelihood to prefer the various products, services, and activities based on text input. The results include the percentiles and raw scores for the personality traits (i.e., dimensions and facets of big five, needs, values, and consumption
