Context-aware Knowledge Management for Socio-Cyber-Physical

Systems: New Trends towards Human-machine Collective

Intelligence

Alexander Smirnov

, Nikolay Shilov

and Andrew Ponomarev

SPIIRAS, 14

Line, 39, St.Petersburg, Russian Federation

Keywords: Socio-Cyber-Physical Systems, Collective Intelligence, Hybrid Systems, Context-aware Knowledge

Management, Ontology-based Systems, Multi-aspect Ontology, Role-based Organization, Dynamic

Motivation.

Abstract: The competitiveness of companies and organizations heavily depends on how they maintain and access

highly decentralized up-to-date information & knowledge coming from various resources located in their

Socio-Cyber-Physical Systems. Such systems tightly integrate heterogeneous resources of the physical

world and IT (cyber) world together with social networking concepts. Context-Aware Knowledge

Management is becoming de facto one of the required business strategies in these systems. Its goal is to

facilitate knowledge transfer and sharing in the context of business structures and activities bound together

with the cultural norms. This keynote presents new trends (including role-based organization, dynamic

motivation mechanisms and multi-aspect ontology) in knowledge management for socio-cyber-physical

systems. Such trends can facilitate creation of innovative IT & HR environments based on human-machine

collective intelligence, where information & knowledge are shared between participants and across

collectives of participants, who can be both people (collective intelligence as the methods used by humans

to act collectively for problem solving) and software services (based on artificial intelligence models). The

keynote considers examples of trends and their implementation experience in a global production company.

https://orcid.org/0000-0001-8364-073X

https://orcid.org/0000-0002-9264-9127

https://orcid.org/0000-0002-9380-5064

1 INTRODUCTION

The concept of Socio-Cyber-Physical System

(SCPS) integrating in real-time physical systems

(e.g., physical production equipment, vehicles,

devices), IT components (e.g., enterprise resource

planning, manufacturing execution systems or other

information systems), and human actors

(organizational roles and stakeholders) at individual

and social network level is becoming more and more

important in understanding modern IT landscape.

Currently, more and more systems (including

“system of systems”) in many areas are recognized

to be socio-cyber-physical, and this spurs on the

research in the area of SCPSs, aimed at creating

coherent tools and methodologies for the SCPSs

development and evolution. Quite a few good

SCPSs examples can be found in modern production

environments, especially those adopting the Industry

4.0 concept.

Advances in the mobility, cloud computing,

crowdsourcing, and big data analytics increase the

number and kinds of networked connections in

business environments, as well as the opportunities

for people and machines to derive unpredictable

value from these connections (Pew Research Center,

2014).

Knowledge management (KM) allowing to

locate knowledge/skill for a task at hand is a crucial

for successful collaboration, in particularly in the

systems with heterogeneous entities (as in SCPSs).

Distributed work in product design, manufacturing,

and supply management projects requires decision

support for the involved parties tailored to the actual

context of these parties (depending on their nature).

Smirnov, A., Shilov, N. and Ponomarev, A.

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence.

DOI: 10.5220/0010171800050017

In Proceedings of the 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2020) - Volume 3: KMIS, pages 5-17

ISBN: 978-989-758-474-9

Network-wise modern SCPSs are based on

integration of a number of networks supported by

the following information technologies (A. Smirnov

& Sandkuhl, 2015):

 Social networks: who knows whom => Virtual

Communities;

 Knowledge networks: who knows what =>

Human & Knowledge Management;

 Information networks: who informs what =>

Internet/Intranet/Extranet/Cloud;

 Work networks: who works where =>

Decision Support based on Crowdsourcing

and Recommendation Systems;

 Competency networks: what is where =>

Knowledge Map;

 Inter-organizational network: organizational

linkages => Semantic-Driven Interoperability.

In general, SCPSs are reconfigurable dynamic

systems; their elements may have variety of possible

states and arrange in dynamically arrange in

problem-centric compositions. This provides an

additional requirement for successful KM in SCPSs.

Namely context-awareness. The context is usually

defined as any information that can be used to

characterize the situation of an entity, where an

entity is a person, place, or object that is considered

relevant to the interaction between a user and an

application, including the user and applications

themselves (Dey, Abowd, & Salber, 2001).

This paper describes some trends in

implementing context-aware KM in SCPSs.

The rest of the paper is organized as follows.

Section 2 describes some important trends in KM in

SCPSs. Section 3 discusses practical application of

these trends in solving KM problems in a production

company. Finally, section 4 presents a design of a

human-machine collective intelligent environment,

which follows these trends and can be used in a

variety of problem domains to effectively solve KM

problems at decision support by human-machine

collectives.

2 MODERN TRENDS IN CAKM

FOR SCPS

This section describes some modern trends in the

context-aware knowledge management (CAKM) for

socio-cyber-physical systems and shows how the

respective emerging technologies can facilitate the

creation of innovative IT&HR environments.

Ontology-based knowledge representation is in

the core of these trends; it is their enabler. The

purpose of ontologies is to represent knowledge

about a certain domain in a machine-readable way.

Ontologies allow to describe, share and process

knowledge considering its syntax along with its

semantics. They are formal conceptualizations of

certain domain of interest that are shared between

different applications (Gruber, 1993; Staab &

Studer, 2009). The ontology describes concepts,

their relationships and axioms thought to exist in the

given domain. They are considered an efficient

mean to solve the interoperability problem. In

particular, ontologies turn out to be effective in

encoding context.

Context model serves to represent the knowledge

about a current situation (the environment

properties, the current problem, as well as states of

the stakeholders).

These models, for instance, are used to reveal

user preferences based on the analysis of the context

representations in conjunction with the implemented

decisions.

2.1 Role-based Organization

Personalized support is important for modern

business applications. As a rule, it is based on

application of the profiling technology. Each user (a

human or an information system) works on a

particular problem or scenario represented via a

context that may be characterized by a particular

customer order, its time, requirements, etc.

Research efforts in the area of information

logistics show information and knowledge needs of

a particular employee depend on his/her tasks and

responsibilities (Lundqvist, 2007). Therefore, in

business applications the idea of personalization

(identification of implicit context of the request) can

be extended with the knowledge of the user’s role.

Besides, it is also the case that representatives of

adjacent (in terms of business process) roles can

have slightly different goals and use different

terminology (even referring to the same concepts).

The idea of the role-based approach is to

consider the workflows and information models

from perspectives of different roles that deal with

them.

Role-based organization for ontology-based KM

assumes the following steps:

1. Structural information about workflows and

the problem domain is collected and described

in the common ontology.

2. User roles are identified and their relevant

parts of the common ontology are defined.

IC3K 2020 - 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management

3. Tasks assigned to the identified roles are

defined.

4. Knowledge required for performing identified

tasks is defined.

5. Based on the identified roles, tasks and

knowledge new knowledge-based workflows

are defined.

6. Corresponding role-based knowledge support

of the workflows is provided based on the

usage of the common ontology and

knowledge / information storages.

This process repeats for each particular role, with

some knowledge being reused by several roles.

The implementation of the approach is described

in Section 3.1.

2.2 Dynamic Motivation Mechanisms

Despite the prevalence of the KM systems aimed at

improving knowledge sharing within the

organization sometimes these KM mechanisms add

responsibilities and activities that have to be done by

employees and that are not seen as important as the

primary (productive) activities. Therefore,

employees may evade using the organized KM

solutions or even feel threatened by organizing their

knowledge in an accessible manner as it might make

them ‘replaceable’. An important task of

management is to establish open and fair corporate

culture that values KM. One of the most important

aspects that have to be considered is aligning

organization and employees goals via employee

motivation (Friedrich, Becker, Kramer, Wirth, &

Schneider, 2020). Especially, dynamic motivation

(the type of motivation that changes within a short

period of time). A good example of dynamic

motivation used by the retailer companies are: the

best sellers boards, scores in the corporate systems

and etc.

The empirical study has proven that dynamic

motivation seems to yield high levels of

engagement, learning, and of performance and

effectiveness in organizational implementation

processes. In addition, dynamic motivation also

seems to positively contribute to collaborative work

and team performance (Ferreira, Araújo, Fernandes,

& Miguel, 2017).

The use of dynamic motivation in some SCPS

relies on answering two questions:

1) How should the participants be motivated

(what rewards are effective)?

2) What software solutions can be used to

define dynamic motivation mechanisms?

The first question is extensively studied in

human resources management area.

The second one is more relevant to IT. There are

two classes of solutions: specialized solutions

(tailored for the particular problem) and generic

solutions.

An example of using specialized solution for

implementing the dynamic motivation approach to

increase the project management efficiency based on

the competency management system is described in

(Smirnov, Kashevnik, et al., 2019). The solution

includes reference and mathematical models of

language expert network, which are used for the

automated assignment of organization’s personnel to

projects. They allow formalizing not only the

individual employees’ skills, but also their

achievements and strengths.

A prominent example of generic solution is

PRINGL language (Scekic, Truong, & Dustdar,

2015) allowing to define motivation policies in an

application independent way and connect to some

information system via an application programming

interface.

2.3 Multi-aspect Ontology

The purpose of ontologies is to represent knowledge

about a certain domain in a machine-readable way.

However, in some complex domains, like Product

Lifecycle Management (PLM), the application of

ontologies is complicated since it has to deal with

interdisciplinary information and knowledge related

to different phases (Shilov, Smirnov, & Ansari,

2020). The terminology and notations used in

various processes are different since they are aimed

at solving tasks of different nature that require

different techniques (Asmae, Souhail, Moukhtar, &

Hussein, 2017; Palmer, Urwin, Young, &

Marilungo, 2017). To a certain extent, this problem

is similar to that of role-based information

representation, where the information and

knowledge have to be presented to different roles in

different views and terminologies.

Some research efforts were aimed for enriching

ontologies with additional information that could

represent additional facts originally described in a

different notation (e.g., semantic annotations (Liao,

Lezoche, Panetto, & Boudjlida, 2016), DAML+OIL

extensions for configuration problem descriptions

(Felfernig, Friedrich, Jannach, Stumptner, & Zanker,

2003), and others). However, this still cannot be an

efficient solution for problems of integrating

information and knowledge from multiple different

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence

notations and terminologies, which is the case for

PLM.

One of the common solutions for multi-domain

systems is having a common ontology at the top and

its extension for specific sub-domains (e.g.,

configuration problem solving). However, it is not

efficient for dynamic domains with large number of

sub-domains, since this would require a continuous

ontology matching and modifications of the

common ontology.

Ontology matching can also be used separately

for establishing links between multiple domain-

specific ontologies. However, manual ontology

matching would require too much time and efforts in

dynamically changing domains and automatic

ontology matching is still not a reliable instrument

since the existing methods deliver high level of

precision only in narrow domains.

The authors of (Lafleur et al., 2016) propose a

model-driven interoperability framework aimed at

supporting relationships between products and

manufacturing equipment. They form a “connection

framework” describing relationships between

different product ontologies maintained in the PLM

system and different ontologies of manufacturing

capabilities managed in the Manufacturing Process

Management system. However, having multiple

ontologies for different tasks is not an efficient

solution for the problem identified either. Since

translating information from one specific ontology

to another assumes a translation between the source

ontology and the common ontology and then

between the target ontology, what eventually will

cause information losses.

Another approach is to preserve the original

domain ontologies and build an additional layer at

the top of them. The authors of (Hagedorn, Smith,

Krishnamurty, & Grosse, 2019) propose to use a

Basic Formal Ontology (BFO) as a top-level

ontology for describing various engineering

domains, and to re-engineer the existing ontologies

so that they would be compliant to it.

Viewing a problem domain from a number of

viewpoints has resulted in appearance of Multi-

Viewpoints Ontology (MVpOnt). In MVpOnt each

viewpoint corresponds to the knowledge

representation model useful for a particular task,

process, or a group of people co-existing in a

common information environment and sharing some

information and knowledge. These viewpoints are

described in a specialized language for the multi-

viewpoint ontologies called MVP-OWL) (Hemam &

Boufaïda, 2011). In 2018 MVP-OWL was extended

with probabilistic reasoning support (Hemam, 2018).

MVP-OWL extends OWL-DL (the complete

description is presented in (Hemam & Boufaïda,

2011)). Firstly, it supports viewpoints that describe

information and knowledge related to a certain task

or process. Secondly, classes and properties are

divided into two groups: local – observed only from

one viewpoint, and global – observed from two or

more viewpoints. The instances can only be local,

however since MVP-OWL supports multi-

instantiation, the instances can exist in several

viewpoints at the same time. Thirdly, the authors

introduce “bridge rules” of four types, which enable

relating concepts from different viewpoints.

This approach is the most suitable for the

problem set since it supports resolving

terminological issues, and also makes it possible to

preserve original formalisms used in existing

ontologies.

3 CAKM IMPLEMENTATION

This section describes several particular

organizational KM problems and how they are

successfully approached by modern solutions

described in Section 2.

The problem at hand is product and knowledge

management in a large automation manufacturing

company. This section integrates results of several

projects carried out by the research team, the paper

authors belong to, together with representatives of

the company.

3.1 Role-based Organization

This approach was implemented in the frame of the

project reported in (Smirnov, Levashova, & Shilov,

2015).

The first step of the approach implementation

was the ontology creation. The resulting ontology

consists of over 1000 classes organized into a four

level taxonomy based on the VDMA (Verband

Deutscher Maschinen – und Anlagenbau, German

Engineering Federation) classification (VDMA.

German Engineering Federation, 2018). Later it was

extended with descriptions of complex products,

their components and compatibility rules.

At the second step, the major roles, whose

workflows were addressed by KM implementation,

have been identified. They included product

manager, product engineer, production manager, and

production engineer.

Then, at steps 3 and 4, their tasks and

knowledge/information needs were analysed. For

IC3K 2020 - 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management

example, the product manager works with customers

and their needs. Since the terminology used by

customers differs from that used by product

engineers, a mapping between the customer needs

and internal product requirements had to be

established.

At steps 5 and 6 the knowledge-based workflows

were defined, and corresponding supporting tools

were built.

The project showed that such approach enabled

implementing KM incrementally, with initiative

coming from employees. E.g., an experimental

knowledge-based support of one workflow could be

implemented for one user role letting the users

estimate its efficiency and convenience. Then,

workflows reusing some knowledge of the

experimental workflow can be added, etc.

Representatives of other roles seeing the

improvements of the implemented knowledge-based

workflows also wish to join and actively participate

in the identification of the knowledge needed for

their workflows and further turning their workflows

into the knowledge-based ones.

3.2 Dynamic Motivation

An example of successfully leveraging dynamic

motivation is automating and facilitating a

translation process involving company employees

from different countries (A. Smirnov, Kashevnik, et

al., 2019). The translation process was implemented

as a distributed network of language experts, and

dynamic motivation was leveraged to incentivize

experts (found by maximization of the global fitness

function) to take part in the translation.

An example of the generated skill tree that is

used to describe expert’s competence profile as well

as task requirements is presented in Fig. 1. The skill

tree for the developed language experts network

consists of three main parts: dictionary, industry

Figure 1: Example of skill tree.

segment, and technical area that describes the

mentioned problem domains.

Every expert is described by a competence

profile. The expert profile contains: information

about the expert, list of competencies, and

professional assessment (global skill level, GSL).

Global skill level is calculated based on a number of

successfully completed tasks this expert performed,

his/her availability estimation for task performing,

estimation of how long the expert works in the

company, qualification of the expert, and rewards

the expert received from the manager.

For the definition of the proofreading task it is

proposed to use the following structure (see Table I).

The task form accessible to the expert includes the

task structure presented in the table as well as the

task discussion interface that allows proofreaders to

exchange their knowledge about it.

Table 1: Proofreading task description.

Name Description

Due Date Date when the task should be performed

Source Language Source language of the term

Target Language Target language of the term

Term Term to be translated

Translation Translation made by translation agency

Task Context

Context that helps an expert to perform the

translation. It includes the project where the

translation will be used, “in sentence

context”, technical area, industry segment,

and etc.

The list of possible motivations used in the

system includes two main groups of motivations:

material and non-material. Every motivation is

specified by budget, value, and monetary benefit as

well as it can be supported globally or only by one

or several local companies.

For example, an expert can be motivated by a

shopping voucher (20 EUR). In this case spent

budget will be 20 EUR as well as monetary benefit

that determines the value of this present for the

expert. At the same time the value of a positive

recommendation to the expert’s boss could be

evaluated as 10 (maximum value) but budget and

monetary benefit is 0, since the company does not

spend money on it.

The system also provides a number of forms for

managing rewards. First, a form to display the list of

rewards assigned to each expert (including date/time

of the assignment) and define new rewards (Fig. 2).

Using this form, a manager can select an expert(s)

and reward. The system shows the left monetary

benefit for each expert in the current year.

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence

Figure 2: Example of the new reward definition.

3.3 Multi-aspect Ontology

Several projects carried out for the same production

company have led to a necessity to share

information and knowledge between several

workflows and departments that do not share the

same terminology. Besides, different tasks of

different workflows required application of different

formalisms what resulted in a necessity of

developing a multi-aspect ontology (A. Smirnov,

Shilov, & Parfenov, 2019). These different views

can be successfully synchronized and matched with

a help of multi-aspect ontology, being a formalized

instrument supporting various processes of the

considered company. For this reason, the ontology

had to cover processes, that were addressed during

development of the information and knowledge

management systems. As an illustrative example for

this paper the aspects of “Product Engineering”,

“Sales”, and “Strategic Planning and Production”

that correspond to different PLM phases have been

selected.

Development of the aspect ontologies can be

done on the basis of any existing methodology of

ontology development, e.g., METHONTOLOGY

(Fernández-López & Gómez-Pérez, 2002). Aspect

ontology can be also built using a different

methodology since the aspects are independent.

When developing an aspect ontology, a reuse of

existing ontologies is beneficial (e.g., typical

subproblems usually already have established

ontologies) though not obligatory.

The illustration of the developed multi-aspect

ontology is given in Fig. 3. The ontology was built

based on the top-level ontology presented in

(Borsato, Estorilio, Cziulik, Ugaya, & Rozenfeld,

2010). Earlier developed ontologies for different

tasks have been used as the aspects (one task

corresponds to one aspect). For the illustration

purposes the aspects with different formalisms have

been selected. Below, each of them is described with

corresponding references.

The first considered aspect Product Engineering

describes the task of definition of a new product and

its features (Oroszi, Jung, Smirnov, Shilov, &

Kashevnik, 2009), which is currently done in the

NOC tool. This aspect is defined in OWL. The goal

of this task is definition of new products and product

families with their possible characteristics by a

product engineer. During this process, the product

engineer has to make sure that the defined products

and characteristics are consistent (the Pellet reasoner

is used for this purpose). The sample classes

presented in the figure include “Product Family”

(high level generalization of products), “Product

Group” (lower level generalization of products, a

subclass of Product Family), “Product” (simple or

modular product, a subclass of Product Group), and

“Feature” (product characteristics, associated with

the class Product).

The second considered aspect is Sales. It

describes the task of defining and using constraints

between product characteristics and product

combinations in an assembly. Definition of the

constraints is done via the CONSys tool by product

manager, and their usage is done in the CONFig tool

by a customer or product/solution managers (A. V.

Smirnov, Shilov, Oroszi, Sinko, & Krebs, 2018). For

the purpose of constraint satisfaction technology

support, the formalism of object-oriented constraint

networks was used. The example classes from this

aspect are “Product” (can be a product or a product

combination), “Parameter” (parameter of a product,

e.g., “mass”, “power”, that can match product

characteristic but it is not always the case), and

“Constraint” (mathematical constraints limiting or

calculating values of product characteristics

depending on other characteristics).

The third presented aspect is Strategic Planning

and Production. The task solved in this aspect is

definition of strategy regarding production classes.

Three classes are considered: “ETO” (engineered to

order, longest lead time), “ATO” (assemble to order,

medium lead time), and “PTO” (pick to order,

shortest lead time) (A. V. Smirnov et al., 2018).

Solving this task is based on pre-defined rules, and,

hence it is defined as a set of classes and production

rules (“if … then …”). Based on these rules the lead

times and production plants for the products are

defined. Example classes of this aspect are

“Production Class” (the superclass for the above

mentioned “ETO”, “ATO”, and “PTO” classes),

“Product”, and “Plant”.

IC3K 2020 - 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management

To sum up the following elements of the multi-

aspect PLM ontology have been defined:

Aspects: Product Engineering, Sales, Strategic

Planning and Production.

Local Classes (by aspect):

Product Engineering aspect: Product Family,

Product Group, Product, Feature.

Sales aspect: Product, Parameter, Constraint.

Strategic Planning and Production aspect:

Product, Production Class, Plant, Rule.

Global level has the following classes: Thing,

Attribute, Product, Dependency, Group, Resource.

To establish connections between aspects and the

global level, bridge rules have been defined. As an

example, the bridge rules of bidirectional inclusion

(symbol ↔





), meaning that two concepts from

different aspects are equal, for the class Product are

presented:

Product ↔





Product

ProductEngineering

;

Product ↔





Product

Sales

;

Product ↔





Product

StrategicPlanningAndProduction

i.e., the concept product in different aspects has

the same meaning.

The resulting ontology made it possible to

establish links between heterogeneous information

models, and, for example, changes made in the

Product Engineering task can be easily reflected in

the Sales.

Figure 3: Multi-aspect ontology for three aspects.

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence

4 CONCEPT OF

HUMAN-MACHINE

COLLECTIVE INTELLIGENCE

ENVIRONMENT

The experience of implementing the above novel

techniques for CAKM in a production environment

and benefits they brought have led to an idea that

they could be applied in a more general way to

create an environment supporting human-machine

collective intelligence.

The problem of human-machine collaboration

and collective intelligence in particular have

attracted attention of researchers in several

perspectives and have posed a number of important

questions (Jennings et al., 2014).

The proposed human-machine collective

intelligence environment is based on the following

foundations:

 One of the established facts about

collaborative work on complex problems is

that it requires certain agent autonomy and

self-organization (Retelny, Bernstein, &

Valentine, 2017).

 To achieve interoperability between human

and software participants, the environment

should support some structured representation

of the discourse contents and/or task

distribution. A good example is the Dicode

project implemented within the framework of

the European FP7-ICT program (Karacapilidis

& Tampakas, 2019), proposing an ontological

presentation of the argumentation process and

a number of visual tools for working with a

formalized set of interrelated arguments.

In this research, however, an environment is built

where heterogeneous agents (human and software)

would be able to collectively decide on the details of

the workflow. Its distinguishing features are:

 The support for self-organization (in contrast

to pre-defined workflows);

 Flexible role-based distribution of

responsibilities;

 The use of ontologies (and, in particular,

multi-aspect ontologies) to support human-

machine interoperability and knowledge

management.

The purpose of this environment is to implement

basic discovery, information exchange and

organization routines to allow agents of different

nature (human and software) to collectively tackle

organizational decision-making problems.

The primary goal is to support cooperation of

relatively short-lived (hours to several days) ad hoc

teams. Another limitation is that the environment is

inherently dedicated to decision support problems.

Therefore, the design is influenced by decision-

making methodologies and the workflow

implemented by a team mostly corresponds to a

typical decision-making process.

There are following principal actors

differentiated by the environment design: end-user

(decision-maker), participant, and service provider

(Fig. 4). End-user (decision-maker) uses the

environment to get help in making a decision.

He/she describes the problem and posts it so that the

problem description is visible to a specified

community. Participant is an active entity (human or

a software service) working on a problem given by

the end-user. Finally, a service provider develops,

integrates to the environment, and supports software

services that can act as participants working on some

problem given by the end user. Service provider is

also responsible for the deployed services, assuaging

the problem of service accountability.

Core entities involved in most of the

environment processes are the problem and the

team. Problem is introduced by an end-user and then

is addressed by the participants’ team. The problem

description has a complex structure and

representation. First of all, it contains information,

specified by the end-user (initial statement), and also

includes all information produced by the team. So,

during team’s activity the problem becomes more

and more detailed. Second, to enable (at least,

partially) an effective interpretation by software

agents the problem description is represented in a

semi-structured way. In particular, machine

readability is achieved via using ontologies. To

facilitate the use of ontologies for people the

environment makes it as implicit as possible by

relying on three techniques:

 Implicit ontological representation of the

structure of problem information.

 Natural language processing. Using advances

in this area it is possible to infer the role of

some information pieces, its relationship with

the goal and/or some line of argumentation

and so on.

 GUI-based nudging participants to encode

problem structure in an ontology-compatible

way.

The environment defines two basic ontologies,

representing different aspects of the collaborative

decision support (Figure 5):

IC3K 2020 - 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management

Figure 4: Principal actors.

 Decision-making ontology. This ontology

defines main concepts that are used during

decision-making (criterion, alternative,

evaluation etc.) and interaction between them.

The ontology is based on the analysis of

existing decision-making methodologies and

has been built to support majority of them.

 Collaboration and coordination ontology. It

defines the concepts used in distributing work

among team members (role, responsibility,

dependency etc.).

The use of above ontologies allows artificial

agents to ‘understand’ the processes taking place in

the team and contribute to them. However, for the

ontology-based decision-support agents, there also

exists a possibility to define an application ontology

and to map it to the decision-making ontology. By

this process some parts of the problem situation

become connected to the general decision-making

terminology.

The way problem information becomes richer

and grows via interaction of agents, to some extent

resemblances to stigmergy (Heylighen, 2016) and

intelligent systems based on the blackboard

interaction principle.

Another already mentioned core entity is the

team. The team in the context of the environment is

defined as a heterogeneous group (consisting of

human participants and software services) working

towards solution of a particular problem. Each

problem has a team dedicated to it. Obviously, a

participant may be a member of several teams, or

not be a member of any team.

Initial team formation is based on the same

principles used in most of the crowdsourcing

platforms and knowledge networks (Ahmad, Battle,

Malkani, & Kamvar, 2011): each participant has a

profile describing key specializations, problem-

solving history, as well as the history of previous

collaborations (with mutual evaluations). There is a

massive list of publications why each of this

components of the profile is necessary and how it

affects the efficiency of teaming. The initiative in

this process is mixed in the sense that a contributor

should send a proposal to the end-user, consisting of

one or more team members (proposal may include

several participants that already have some positive

experience of working together), and end-user has to

collect the initial team. However, decisions of the

both parties – participants and end-users – are

assisted by environment. The participants may

choose to receive recommendations if some problem

touching his/her area of competences is posted. On

the other hand end-users may explore the description

and history of all the participants mentioned in the

proposals.

Due to much uncertainty typically associated

with decision-making, it is often the case that during

the work on the problem, the team understands that

it lacks some competencies or resources. Therefore,

the team may create a new resource requirements,

that are registered in the environment and resolved

in a manner, similar to the initial team formation

process (participants have to actively apply for the

positions in the team, however, both sides are

assisted by the environment mechanisms).

It should be noted, that it does not fully apply to

the software participants (services). As the

throughput of software services is not as limited as

the throughput of humans, and the execution is

relatively cheap, software services are passively

connected to any team and by the mechanisms of the

environment (ontology-based publish-subscribe) are

watching the processes taking place with the

problem. There are two states a software service can

be in w.r.t. the team: dormant and active. Initially,

all services are in the dormant state and are waiting

for specific conditions during the problem-solving.

If these conditions defined by a particular service are

met, the service tries to activate, describing its

purpose and terms of use. If the team agrees that the

service is useful for the problem, the service is

allowed to activate (change state to active) and

become a member of the team. Otherwise, the

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence

Figure 5: Conceptual model of the environment.

service remains dormant. Active service may also be

transferred to the dormant state by a decision of the

team. Besides, the services can be accessed via a

service catalogue and activated manually by team

members.

Active services can be used by the team

members. The mechanics of their usage depends on

the service’s kind. There are two main types of

services:

 Problem-solving service;

 External tool and database access service.

Problem-solving service accesses the problem

information described in the form of ontology and

natural text, and can actively add information pieces

to it. An example of such service is a statistics-based

question answering service – if it detects a question

about some facts (e.g., “How many people die from

tuberculosis in the World in one year?”) and can

answer it in some form, it adds an answer to the

question. Another example is a service that derives

from the problem information a current set of

alternatives and their evaluations, builds a Pareto

optimal set and adds it to the problem information.

External tool and database access services in

their activated form only provide an access to a

specified resource. For example, if the team needs

an epidemic database, it can activate the service that

grants access to this database and use it for queries.

Simultaneously two processes take place when

team works on a problem: solution preparation and

decision support (re)organization. Both of these

processes are supported by mechanisms provided by

the environment. Solution preparation is main

productive process, during which problem is

enriched with new information and artefacts created

by team members. The result of this process is fully

detailed description of a problem situation, weighted

alternatives and their estimated consequences –

accepted by the end-user. Decision support

(re)organization process represents all the activities

aimed at planning and organization of team work

(e.g., deciding whether additional resources are

required, assigning team member responsibilities,

setting task deadlines and identifying new tasks to

be solved in order to reach the goals of the whole

process).

Several ontologies used to describe the current

problem state are connected by the multi-aspect

ontology approach. Two main aspects used in

describing the problem are decision support aspect

and domain aspect. For example, if the environment

is used in a smart tourism scenario to select a tourist

itinerary, then possible alternatives to consider and

evaluate are tourist itineraries. Therefore, class

Alternative of the decision-making aspect in this

problem setting is connected with the Itinerary

IC3K 2020 - 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management

concept of the domain aspect. Further, evaluation of

the itineraries done by the team can be interpreted as

evaluations of the alternatives which is an essential

step in making a decision. It should be noted, that

mutual mapping of the aspects is realized in the

context of the problem (in location problems

Alternative is mapped to geographic point, etc.).

By providing a set of mechanisms (maintaining

the problem state, discourse logics, team formation

processes etc.) the environment supports the

activities of the human-machine team. One

important specific mechanism provided by the

environment is soft guidance by offering situation-

specific cooperation patterns to the team. In this

sense, the environment plays the role similar to the

facilitator in group decision support systems. These

recommendations are based on the number of

identified collaboration patterns: generate, reduce,

clarify, organize, evaluate, build consensus. These

patterns form basic activities performed by members

of the team. Sometimes, current activity is defined

explicitly during decision support (re)organization

process (e.g., there might be an alternative

generation activity). In other cases, pattern can be

recognized by certain structure in the ontological

description of the problem state (e.g., if two or more

participants offer different estimations for the same

alternative, then these estimations should be

reconciled during consensus building). An important

role of these patterns is that they allow to structure

the team activities (w.r.t. the goal) and tie existing

methods to these activities. For example, if it has

been recognized, that the team needs to build

consensus on the set of the alternatives estimation, a

number of methods for building consensus could be

recommended.

It can be seen, that the design of the environment

is heavily affected by the discussed trends. First of

all, ontology-based context modeling and

specialization are at the core of the problem

representation. The problem situation and all the

artefacts are represented in the form of ontology,

which allows to achieve interoperability between

human and software participants. Role-based

organization and multi-aspect ontologies are used to

help to reconcile different aspects of the decision

support (e.g., domain structure vs. process structure),

besides, role-based organization is used also in the

foundational layer, because every decision-making

process in a coarse decomposition can be viewed as

an interaction of different roles (project leader, data

analyst, domain expert, etc.), and the environment

supports the definition of the team via set of roles.

Finally, dynamic motivation mechanisms play a role

in process planning and team recruiting, because

reward sharing is an important aspect of process

definition.

5 CONCLUSIONS

The paper discusses three modern trends in context-

aware knowledge management for socio-cyber-

physical systems. In particular:

 role-based organization,

 dynamic motivation mechanisms, and

 multi-aspect ontology.

Each of these trends (or their combination) can

be implemented in a variety of systems, improving

the effectiveness of knowledge eliciting, storing, and

utilization, which can have a major positive impact

on the effectiveness of the whole system

(organization).

The paper also introduces a concept of human-

machine collective intelligence environment, making

use of all these trends.

ACKNOWLEDGEMENTS

The research is partially funded by the Russian State

Research, project 0073-2019-0005. The research on

the human-machine collective intelligence for

decision support is funded by the Russian Science

Foundation, project 19-11-00126.

REFERENCES

Ahmad, S., Battle, A., Malkani, Z., & Kamvar, S. (2011).

The jabberwocky programming environment for

structured social computing. Proceedings of the 24th

Annual ACM Symposium on User Interface Software

and Technology - UIST ’11, 53–64.

https://doi.org/10.1145/2047196.2047203

Asmae, A., Souhail, S., Moukhtar, Z. El, & Hussein, B.

(2017). Using ontologies for the integration of

information systems dedicated to product (CFAO,

PLM…) and those of systems monitoring (ERP,

MES). 2017 International Colloquium on Logistics

and Supply Chain Management (LOGISTIQUA), 59–

64.

https://doi.org/10.1109/LOGISTIQUA.2017.7962874

Borsato, M., Estorilio, C. C. A., Cziulik, C., Ugaya, C. M.

L., & Rozenfeld, H. (2010). An ontology building

approach for knowledge sharing in product lifecycle

management. International Journal of Business and

Systems Research, 4(3), 278.

https://doi.org/10.1504/IJBSR.2010.032951

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence

Dey, A. K., Abowd, G. D., & Salber, D. (2001). A

Conceptual Framework and a Toolkit for Supporting

the Rapid Prototyping of Context-Aware Applications.

Human–Computer Interaction, 16(2–4), 97–166.

https://doi.org/10.1207/S15327051HCI16234_02

Felfernig, A., Friedrich, G., Jannach, D., Stumptner, M., &

Zanker, M. (2003). Configuration knowledge

representations for Semantic Web applications.

Artificial Intelligence for Engineering Design,

Analysis and Manufacturing, 17(01), 31–50.

https://doi.org/10.1017/S0890060403171041

Fernández-López, M., & Gómez-Pérez, A. (2002).

Overview and analysis of methodologies for building

ontologies. The Knowledge Engineering Review,

17(2), 129–156.

https://doi.org/10.1017/S0269888902000462

Ferreira, A. T., Araújo, A. M., Fernandes, S., & Miguel, I.

C. (2017). Gamification in the Workplace: A

Systematic Literature Review. Recent Advances in

Information Systems and Technologies. Advances in

Intelligent Systems and Computing, 571, 283–292.

https://doi.org/10.1007/978-3-319-56541-5_29

Friedrich, J., Becker, M., Kramer, F., Wirth, M., &

Schneider, M. (2020). Incentive design and

gamification for knowledge management. Journal of

Business Research, 106, 341–352.

https://doi.org/10.1016/j.jbusres.2019.02.009

Gruber, T. R. (1993). A translation approach to portable

ontology specifications. Knowledge Acquisition, 5(2),

199–220. https://doi.org/10.1006/knac.1993.1008

Hagedorn, T. J., Smith, B., Krishnamurty, S., & Grosse, I.

(2019). Interoperability of disparate engineering

domain ontologies using basic formal ontology.

Journal of Engineering Design, 1–30.

https://doi.org/10.1080/09544828.2019.1630805

Hemam, M. (2018). An Extension of the Ontology Web

Language with Multi-Viewpoints and Probabilistic

Reasoning. International Journal of Advanced

Intelligence Paradigms, 10(1), 1.

https://doi.org/10.1504/IJAIP.2018.10003857

Hemam, M., & Boufaïda, Z. (2011). MVP-OWL: a multi-

viewpoints ontology language for the Semantic Web.

International Journal of Reasoning-Based Intelligent

Systems, 3(3/4), 147.

https://doi.org/10.1504/IJRIS.2011.043539

Heylighen, F. (2016). Stigmergy as a universal

coordination mechanism I: Definition and

components. Cognitive Systems Research, 38, 4–13.

https://doi.org/10.1016/j.cogsys.2015.12.002

Jennings, N. R., Moreau, L., Nicholson, D., Ramchurn, S.,

Roberts, S., Rodden, T., & Rogers, A. (2014). Human-

agent collectives. Communications of the ACM,

57(12), 80–88. https://doi.org/10.1145/2629559

Karacapilidis, N., & Tampakas, V. (2019). On the

Exploitation of Collaborative Argumentation

Structures for Inducing Reasoning Behavior.

Proceedings of the 18th International Conference on

IEEE/Internet

Lafleur, M., Terkaj, W., Belkadi, F., Urgo, M., Bernard,

A., & Colledani, M. (2016). An Onto-Based

Interoperability Framework for the Connection of

PLM and Production Capability Tools. PLM 2016:

Product Lifecycle Management for Digital

Transformation of Industries, 134–145.

https://doi.org/10.1007/978-3-319-54660-5_13

Liao, Y., Lezoche, M., Panetto, H., & Boudjlida, N.

(2016). Semantic annotations for semantic

interoperability in a product lifecycle management

context. International Journal of Production

Research, 54(18), 5534–5553.

https://doi.org/10.1080/00207543.2016.1165875

Lundqvist, M. (2007). Information Demand and Use:

Improving Information Flow within Small-scale

Business Contexts.

Oroszi, A., Jung, T., Smirnov, A., Shilov, N., &

Kashevnik, A. (2009). Ontology-driven codification

for discrete and modular products. International

Journal of Product Development, 8(2), 162–177.

https://doi.org/10.1504/IJPD.2009.024186

Palmer, C., Urwin, E. N., Young, R. I. M., & Marilungo,

E. (2017). A reference ontology approach to support

global product-service production. International

Journal of Product Lifecycle Management, 10(1), 86.

https://doi.org/10.1504/IJPLM.2017.083003

Pew Research Center. (2014). Digital life in 2025.

Retrieved 18 August 2020, from

http://www.pewinternet.org/files/2014/03/PIP_Report

_Future_of_the_Internet_Predictions_031114.pdf

Retelny, D., Bernstein, M. S., & Valentine, M. A. (2017).

No Workflow Can Ever Be Enough: How

Crowdsourcing Workflows Constrain Complex Work.

Proceedings of the ACM on Human-Computer

Interaction, 1(2), Article 89.

https://doi.org/10.1145/3134724

Scekic, O., Truong, H.-L., & Dustdar, S. (2015). PRINGL

– A domain-specific language for incentive

management in crowdsourcing. Computer Networks,

90, 14–33.

https://doi.org/10.1016/j.comnet.2015.05.019

Shilov, N., Smirnov, A., & Ansari, F. (2020). Ontologies

in Smart Manufacturing: Approaches and Research

Framework. 2020 26th Conference of Open

Innovations Association (FRUCT), 408–414.

https://doi.org/10.23919/FRUCT48808.2020.9087396

Smirnov, A., Kashevnik, A., Petrov, M., Shilov, N.,

Schafer, T., & Jung, T. (2019). Competence-Based

Language Expert Network for Translation Business

Process Management. 2019 25th Conference of Open

Innovations Association (FRUCT), 279–284.

https://doi.org/10.23919/FRUCT48121.2019.8981515

Smirnov, A., Levashova, T., & Shilov, N. (2015). Role-

driven context-based decision support: Approach,

implementation and lessons learned. In

Communications in Computer and Information

Science (Vol. 553). https://doi.org/10.1007/978-3-319-

25840-9_32

Smirnov, A., & Sandkuhl, K. (2015). Context-Oriented

Knowledge Management for Decision Support in

Business Networks: Modern Requirements and

Challenges. BIR 2015 Workshops, Vol. 1420.

IC3K 2020 - 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management

Smirnov, A., Shilov, N., & Parfenov, V. (2019). Building

a Multi-aspect Ontology for Semantic Interoperability

in PLM. In Product Lifecycle Management in the

Digital Twin Era. IFIP Advances in Information and

Communication Technology (Vol. 565, pp. 107–115).

https://doi.org/10.1007/978-3-030-42250-9_10

Smirnov, A. V., Shilov, N., Oroszi, A., Sinko, M., &

Krebs, T. (2018). Changing information management

for product-service system engineering: Customer-

oriented strategies and lessons learned. International

Journal of Product Lifecycle Management, 11(1), 1–

18. https://doi.org/10.1504/IJPLM.2018.091647

Staab, S., & Studer, R. (Eds.). (2009). Handbook on

Ontologies. https://doi.org/10.1007/978-3-540-92673-

VDMA. German Engineering Federation. (2018).

Retrieved from www.vdma.org/en_GB/

Context-aware Knowledge Management for Socio-Cyber-Physical Systems: New Trends towards Human-machine Collective Intelligence