CogBPMN: Representing Human-computer Symbiosis

in the Cognitive Era

Juliana Jansen Ferreira, Viviane Torres da Silva, Raphael Melo Thiago, Leonardo Guerreiro Azevedo

and Renato F. de G. Cerqueira

IBM Research Brazil, Brazil

Keywords:

Cognitive Computing, Cognitive Systems, Business Process Notation, BPMN, Cognitive BPMN,

Human-computer Symbiosis.

Abstract:

The human-computer symbiosis is a core principle of Cognitive Computing where humans and computers are

coupled very tightly, and the resulting partnership presents new ways for the human brain to think and com-

puters to process data. Business Process Management (BPM) provides methods and tools to represent, review,

and discuss business domains, considering their knowledge, context, people, computer systems, and so on.

Such methods and tools will be affected by advances in Cognitive Computing. Business Process Modeling

notations need to support discussion and representation of human-computer symbiosis in any given organi-

zational context. We propose CogBPMN, a set of cognitive recommendation subprocesses types that can be

used to represent human-computer symbiosis in business process models. With CogBPMN, business stake-

holders and Cognitive Computing specialists can understand how business processes can thrive by considering

cognitive empowerment in organizations’ core processes. We discuss the proposed cognitive subprocesses in

a medical domain use case.

1 INTRODUCTION

In the past, Cognitive Computing aimed to develop

a coherent, uniﬁed, universal mechanism inspired

by the mind’s capabilities (Modha et al., 2011). It

focused on investigating the development of self-

learning systems, which naturally interact with hu-

mans in complex environments, and are capable of

adapting themselves to context. More recently, the

idea of a human-computer symbiosis is gaining mo-

mentum in the Cognitive Computing research (Kelly,

2015) where humans and computers collaborate, us-

ing their unique and powerful capabilities, to build an

environment where knowledge is created and evolves

considering environment events. Computers bring

their capability to deal, in different ways, with large

sets of data, which is impractical for the human brain.

On the other hand, humans provide their capability to

judge and make decisions, to assess situations consid-

ering their intuition and all kinds of knowledge (struc-

tured, unstructured, subjective, etc.), and their abil-

ity to innovate in a given domain. Understanding this

human-computer symbiosis is the key to the applica-

tion of proper Cognitive Computing resources in any

business domain.

The relationship between humans and comput-

ers is intricately established in any business and so-

ciety. However, to understand and discuss it, con-

sidering the advantages of Cognitive Computing re-

sources to the modeled business, we need to rep-

resent that relationship. Business Process Manage-

ment (BPM) practices already present process mod-

eling notations to represent business domains. BPM

also investigates the bridge between business context

and software systems, considering business process

models as a source of early software system’s re-

quirements (Alotaibi and Liu, 2017). Therefore, busi-

ness process models are a representation that can be

used to express the human-computer symbiosis nec-

essary to deﬁne artifacts for the development of cog-

nitive systems. However, BPM still requires a “cog-

nitive layer” above its practices to address the evolu-

tion and adaptation to the Cognitive Era. Cognitive

BPM, coined in (Nezhad and Akkiraju, 2014), refers

to a new paradigm in BPM, which encompasses all

BPM contexts and aspects of its ecosystem that are

impacted and enabled by Cognitive Computing tech-

nologies (Hull and Nezhad, 2016).

Any business process, from transaction-intensive

to knowledge-intensive (Di Ciccio et al., 2015),

850

Ferreira, J., Torres da Silva, V., Thiago, R., Azevedo, L. and Cerqueira, R.

CogBPMN: Representing Human-computer Symbiosis in the Cognitive Era.

DOI: 10.5220/0009787408500858

In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 2, pages 850-858

ISBN: 978-989-758-423-7

has the potential for cognitive empowerment through

human-computer symbiosis. Several organizations

have their business processes designed with Business

Process Model and Notation (BPMN) - the de-facto

standard for business process modeling (Chinosi and

Trombetta, 2012). BPMN already deals with human-

computer relation, e.g., the user representation sup-

ported by system task characterizes a human per-

forming a task with the assistance of a software sys-

tem (OMG, 2011). However, we believe that BPMN

needs resources to support the discussion, representa-

tion, and speciﬁcation of human-computer symbiosis

necessary to develop cognitive systems.

The relation between humans and computers deals

with other aspects of Cognitive Computing. For ex-

ample, machine learning techniques support various

human tasks involved in interpretative-intensive ac-

tivities (Kelly, 2015). Some of those tasks, as de-

picted in (KPMG, 2015), are the identiﬁcation of con-

cepts in massive data, ranking web searches, pattern

recognition, trend ﬁnding, and sentiment analysis on

social media. With the immediate dissemination of

such technologies, the relationship between humans

and AI algorithms in daily tasks is undoubtedly be-

coming more transparent. The relationship between

humans and computers in these cognitive scenarios

is symbiotic: the quality of such cognitive systems

increases because of the interaction itself. In this

paper, we propose CogBPMN – four recommenda-

tion cognitive subprocesses types to represent human-

computer symbiosis in business process models. The

subprocesses indicate activities to represent actions

for recommendation, creating evidence from the rea-

soning used to reach recommendations, and getting

user feedback. Besides, the subprocesses suggest rep-

resentations for the involved data and knowledge. We

exemplify the use of CogBPMN, designing a business

process of the medical domain scenario.

Our main motivations to propose CogBPMN are:

(i) explore the representation of processes focusing on

the modeling of Cognitive Computing activities; and,

(ii) use BPMN to allow the analysis and (re)design of

organization models containing such processes. Rep-

resent human-computer symbiosis in a business pro-

cess that can be used by stakeholders and Cognitive

Computing specialists to (re)analyze and (re)think

processes and decide where the Cognitive Computing

resources can make a strategical difference for busi-

nesses. In the current version of CogBPMN presented

in this paper, we have chosen to focus on the model-

ing of recommendation processes (Rich, 1979), which

is a classic standard process coming from experts and

recommendation systems.

The remainder of this work is divided as follows.

Section 2 presents the background. Section 3 presents

our proposal. Section 4 presents an example of the

use of the proposal. Finally, Section 5 concludes and

presents future work.

2 BACKGROUND

Cognitive Computing refers to software (cognitive)

systems that learn at scale from their interaction with

humans and the environment. One of the deﬁning

characteristics of cognitive systems is that they are

capable of improving their performance over time by

leveraging its understanding of their users and ap-

plication domain. Some requirements are: (i) inter-

actions should be (at least partially) recorded; and,

(ii) models should learn from previous interactions.

The research and development of cognitive systems

involve multiple computer science areas of exper-

tise. There must be a constant technical collabora-

tion from different professionals to tackle the chal-

lenges of building a cognitive system. The areas in-

volved depend on the problems that the cognitive sys-

tem aims to solve. At least HCI (Human-Computer

Interaction) experts (UX designers and HCI profes-

sionals) and developers need to work closely together

to build a cognitive system. Designers and HCI pro-

fessionals, working together with industry experts,

place the new system in real scenarios; and developers

and machine learning experts deﬁne the algorithmic

strategy and datasets for solving the scenario prob-

lems. This multidisciplinary technical approach ap-

plied to Cognitive Computing differs from other cur-

rent approaches of Artiﬁcial Intelligence (AI) that fo-

cus on algorithmic accuracy and not on the people that

work with the system (Kelly, 2015). Cognitive sys-

tems allow humans and computers to collaborate and

produce better results, each one bringing their supe-

rior skills to the partnership: computers with rational

and analytics and people with intuition, empathy, and

judgment. That is why this is such a multidisciplinary

approach (Kelly III and Hamm, 2013). HCI (Human-

Computer Interaction) is an important research area to

investigate and explore human-computer symbiosis.

HCI and AI expertise have been combined for some

time (Grudin, 2009). The combination of HCI and

AI research has been a concern of major tech com-

panies like Google

, which launched an initiative to

study and redesign the ways people interact with AI

systems (Holbrook, 2017).

We argue that a cognitive system learns when-

ever the output of a given model evolves due to

https://ai.google/pair/

CogBPMN: Representing Human-computer Symbiosis in the Cognitive Era

851

changes in the knowledge base. Therefore, one re-

quirement is that models, in principle, should use the

acquired knowledge in their predictions. This deﬁni-

tion of learning is loosely based on (Mitchell and et al,

1997). In particular, we relax the performance mea-

surement requirement, since evaluating “how good is

the human-computer symbiosis” is a non-trivial task:

A computer program is said to learn from experience

E with respect to some class of tasks T and perfor-

mance measure P, if its performance at tasks in T, as

measured by P, improves with experience E.

Recommendation and Expert systems are refer-

ences in AI for Cognitive Computing approaches.

Recommendation, or recommender systems, are soft-

ware tools and techniques that provide suggestions to

support users’ decisions (Adomavicius and Tuzhilin,

2005). Recommendation systems deal with the prob-

lem of estimating ratings for current items, usually

based on the ratings given by the user to other items

and/or by other users to the same items. Expert

systems reconstruct the expertise and reasoning ca-

pabilities of qualiﬁed specialists within limited do-

mains. The underlying assumption is that experts con-

struct their solutions from single pieces of knowledge,

which they select and apply in a proper sequence.

Hence, expert systems require detailed information

about the domain and the strategies for applying this

knowledge to problem-solving. Those systems simu-

late problem-solving tasks over static representations

of some knowledge domain (Kidd, 2012).

Although recommender systems include concepts

of cognitive science (Rich, 1979), they handle rec-

ommendation problems that explicitly rely on rat-

ing structures. On the other hand, Cognitive Com-

puting enables “knowledge acquisition at scale”, i.e.,

deeper than ratings of items, using emerging meth-

ods in natural language understanding and machine

learning techniques (Hull and Nezhad, 2016). Expert

systems, different from cognitive systems, do not aim

for human-computer symbiosis, but to emulate human

thinking and problem-solving abilities (Kidd, 2012).

The discussion, mapping, modeling, and represen-

tation of business scenarios are rich topics to be devel-

oped by the BPM research area. Those topics are re-

lated to the research of Cognitively-enable BPM (Hull

and Nezhad, 2016), and referenced in our work as

Cognitive BPM. Particularly those scenarios where

people need to harvest insights from vast quantities

of data to understand complex situations, make accu-

rate predictions, and anticipate the unintended con-

sequences of actions and other human-centered pro-

cesses.

There are also important concepts and key

abstractions for Cognitive BPM and Knowledge-

intensive Processes (KiP’s) research (Di Ciccio et al.,

2015)(Hull and Nezhad, 2016)(Netto et al., 2013).

Cognitive Computing presents the possibility of

“knowledge at scale” (Hull and Nezhad, 2016). KiP’s

process representation deals with the life-cycle of

knowledge in business processes. In that way, large

amounts of knowledge relevant to a process instance

can be considered as input for new process’ instances

and the model itself. Moreover, cognitive potential

can be present in structured workﬂow processes to

unstructured and knowledge-intensive ones (Hull and

Nezhad, 2016). Also, the practice of software sys-

tems early requirement derivation from business pro-

cess models (Alotaibi and Liu, 2017) can be applied

for cognitive systems’ speciﬁcations.

BPM has been used in different industries for over

a decade (Van der Aalst, 2013). Hence, there is a large

amount of pre-existing knowledge legacy of business

processes models in organizations. The knowledge

presented in those business models and their instances

is a crucial asset for each organization that wants to

advance to the Cognitive Era. Questions like “What

and how can we learn through discussing and revising

old business processes in the light of Cognitive Com-

puting technologies?” and “What can happen to trans-

actional processes once there is a technology to han-

dle more volume of data and unstructured data?” are

going to guide the Cognitive BPM research agenda.

BPM activities, like mapping, modeling, and analy-

sis of processes, can gain with Cognitive Computing

technology abilities to explore and get insights from

large amounts of business process data.

3 CogBPMN

This section describes CogBPMN, our proposal for

representing cognitive subprocesses using BPMN.

We focus on subprocess types that include recom-

mendations, evidence that support those recommen-

dations, and the handling of user feedback. The rec-

ommendation subprocesses are a guide since they

contain activities that should be used when model-

ing a system that provides recommendations. We

also present an abstract view about how to perform

those activities, although it is not the focus of this

work. Our current proposal of CogBPMN comprises

four subprocesses types: (i) Recommendation sub-

process; (ii) Recommendation with evidence subpro-

cess; (iii) Recommendation with feedback subpro-

cess; and, (iv) Recommendation with feedback and

evidence subprocess.

Section 4 presents an example of the proposal for

a cognitive application in the medical domain. In the

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

852

following, we present the proposed BPMN four sub-

processes types, and their abstract representation us-

ing BPMN subprocesses.

We decide to explore BPMN speciﬁcation (OMG,

2011), which is the de-facto standard modeling lan-

guage for business process modeling (Chinosi and

Trombetta, 2012), to identify deﬁnitions and visual

representations that could support our proposal for

CogBPMN. Table 1 presents the elements and corre-

sponding icons we are using in this work.

Table 1: BPMN elements used in this work.

Element Icon

Service task

User task

Data store

Data object

We selected service and user tasks to represent advi-

sor and user’s tasks, respectively. A service task is

a task that uses some sort of service ((OMG, 2011),

pp.158). Its deﬁnition already covers our need to

represent a task performed by an algorithm or ser-

vice. We choose the user task element to represent

tasks where humans execute an activity with the ad-

visor’s aid. A user task is a typical workﬂow task

where a human performs the task with the assistance

of a software application ((OMG, 2011), pp.163). In

CogBPMN, the software application is the advisor for

the represented subprocess.

We used the data store element to represent a

generic knowledge base in the main process and also

the speciﬁc data stores in the cognitive subprocesses.

A data store provides mechanisms to retrieve or up-

date stored information that will persist beyond the

scope of the process ((OMG, 2011), pp.208). We

created a speciﬁc data store called process instances’

data store, which represents the data of the previously

executed process’ instances. The cognitive subpro-

cess has the potential capacity to learn from its exe-

cution and the main process execution. We used the

data object element ((OMG, 2011), pp.205-206) to

represent speciﬁc instances data, and collection of in-

stances data. Those elements are present in the main

process model and the subprocesses.

BPMN is a ﬂexible notation that allows its users

to add more semantics to its elements, like addressing

different meanings to the same element represented

in different colors. For CogBPMN, there are speciﬁc

demands that the notation speciﬁcation needs to ad-

dress. For example, the data store for process’ in-

stances usually needs to be handled on cognitive sub-

process’ new instances. Data stores remain generic

and central for cognitive processes, representing data

and knowledge generated and consumed throughout

the process’ instances. We argue that CogBPMN

speciﬁcation is a conservative extension (Turski and

Maibaum, 1987) of BPMN to represent speciﬁc activ-

ities that deﬁne a pattern, as a specialized type of sub-

process like Transaction, Ad-Hoc and Event Subpro-

cesses ((OMG, 2011), pp.173-183). We are propos-

ing a visual representation that does not interfere with

the model notation speciﬁcation but makes explicit

and enriches the representation for identifying and

discussing when the human-computer symbiosis hap-

pens in business processes.

3.1 Cognitive Recommendation

Subprocesses

We present in this section, the four Cognitive Recom-

mendation subprocesses. They are abstract represen-

tations, corresponding to suggestions of activities to

be modeled in each case, i.e., the process modeler de-

signs those activities according to the scenario being

modeled.

3.1.1 Recommendation Subprocess

Figure 1 illustrates an abstract representation of the

Recommendation subprocess. This subprocess is ex-

tended by the other three by deﬁning new activities

and other elements. It is the simplest proposed sub-

process.

Figure 1: Recommendation subprocess abstract representa-

tion.

The ﬁrst activity represents the advisor analyzing the

knowledge base (KB) of the application to create hy-

potheses. The second activity represents the advisor

generating the recommendations and storing such rec-

ommendations in (another) KB. These two activities

can be merged into one activity if necessary, repre-

senting the analysis and the generation of recommen-

dations. The third activity represents the actions of

the user monitored by the advisor. This action could

CogBPMN: Representing Human-computer Symbiosis in the Cognitive Era

853

represent, for instance, the selection of an option from

the list of options recommended by the advisor. They

do not need to be executed immediately after the ﬁrst

two actions.

To provide recommendations to a user, the cogni-

tive advisor analyzes available knowledge related to

the application domain, the process being executed,

and the process’ instances already executed. Besides,

the advisor can learn about the recommendations pro-

vided and the actions carried out by the user. For ex-

ample, the advisor evaluates if the user has followed

or not the recommendations and uses such informa-

tion when making new recommendations. This ex-

ample is a kind of implicit feedback produced by the

user for the recommendations provided by the advi-

sor. Some examples are:

• Using the history of actions executed by the user

and his peers, the advisor recommends the next

actions to be executed;

• Using previous recommendations and the most se-

lected options in a speciﬁc context, the advisor

suggests the ones the user may choose.

In both cases above, the advisor monitors the actions

performed by the user to learn about its recommenda-

tions.

3.1.2 Recommendation with Evidence

Subprocess

Figure 2 presents an abstract representation of Recom-

mendation with evidence subprocess. It extends the

Recommendation subprocess by making available to

the user the evidence that supported each generated

recommendation. Such evidence is relevant to the

user to understand why the advisor has provided the

recommendations. Recommendations and evidence

are stored along with the actions taken by the user.

The advisor can use these data to learn and make more

precise recommendations. Examples of this type of

process are:

• Besides recommending the next action, the advi-

sor shows the user evidence like the history of pre-

vious interactions executed by him and by others;

• Besides suggesting options to the user, the advisor

shows evidence like recommended and selected

options in past similar contexts.

3.1.3 Recommendation with Feedback

Subprocess

Figure 3 presents an abstract representation of Rec-

ommendation with feedback subprocess. It extends

the Recommendation subprocess by allowing the user

Figure 2: Recommendation with evidence subprocess ab-

stract representation.

to provide feedback about the advisor’s recommenda-

tions. This kind of feedback is explicit, different from

the implicit feedback already present in the Recom-

mendation subprocess. When given explicit feedback,

the user informs if she/he likes or dislikes a recom-

mendation. The advisor stores the recommendations,

actions executed by the user, and the explicit feed-

back. It uses such knowledge to learn and to make

improved recommendations. Examples of this type of

process are:

• The user likes the recommended action and fol-

lows the recommendation;

• The user likes some of the recommended options

and dislikes others;

• The user does not follow any of the recommended

options.

Figure 3: Recommendation with feedback subprocess ab-

stract representation.

3.1.4 Recommendation with Feedback and

Evidence Subprocess

Figure 4 presents an abstract representation of Recom-

mendation with feedback and evidence subprocess. It

combines the Recommendation with evidence and the

Recommendation with feedback subprocesses by al-

lowing the user to provide feedback not only about the

recommendations made by the advisor but also about

the evidence used to support them.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

854

Figure 4: Recommendation with feedback and evidence

subprocess abstract representation.

The advisor should store the recommendations, evi-

dence, actions executed by the user, and also the feed-

back to provide a more precise recommendation. The

feedback about evidence is essential, e.g., to guide the

advisor on the sources (knowledge-based, documents,

etc.) it uses in its analyses. Some examples of Rec-

ommendation with feedback and evidence subprocess

are:

• The user likes the recommended actions and also

agrees that the previous actions executed in simi-

lar contexts are the correct evidence to be used;

• The user wants the recommended options but does

not agree with the evidence. The user explains the

context considered by the advisor is not similar to

the context of the execution.

The Recommendation with feedback and evidence

subprocess is the most complete proposed cognitive

recommendation subprocess. The other subprocesses

are simpliﬁcations of this one. The ﬁrst and sec-

ond activities (Analyze Knowledge Base and Generate

Recommendation) are present in all recommendation

subprocesses. To provide recommendations to a user,

the cognitive advisor analyzes available knowledge

bases (KBs) of the application domain and the pro-

cess’ instances already executed. Besides, the advisor

can learn from previous recommendations and the ac-

tions carried out by the user stored in the process’ in-

stances KB. The advisor evaluates if the user has fol-

lowed or not the recommendations and uses such in-

formation when making new recommendations. The

information about the user’s decisions for the recom-

mendations is a kind of implicit feedback used by the

advisor.

In the feedback, the user informs if she/he likes or

dislikes the recommendation, and if she/he agrees or

disagrees with the evidence used to support the rec-

ommendation. In both cases, the user should also

be able to include comments (e.g., expressed in nat-

ural language) about the recommendations and evi-

dence. For instance, on the one hand, the user may

like the recommendations but understands that the ev-

idence supporting the recommendations are not ade-

quate. In such a case, the advisor has provided the

right recommendations based on the wrong evidence.

On the other hand, the user may dislike the recom-

mendations, but agree with the evidence. In this case,

the advisor has provided the wrong recommendation

based on accurate evidence.

The advisor should store the recommendations,

evidence, actions executed by the user, and the feed-

back to provide a more precise recommendation. The

feedback about evidence is essential, e.g., to guide

the advisor about the sources (knowledge-based, doc-

uments, etc.) it uses in its analyses.

4 EXAMPLE OF MODEL USING

CogBPMN

This section presents an example using our pro-

posal of cognitive subprocesses types (CogBPMN) to

model the business process that represents the Doc-

tor Oncology scenario (Figure 5). This scenario il-

lustrates the use of IBM Watson for Oncology

, the

AI technology from IBM, which helps physicians

quickly identify critical information in a patient”’s

medical record, surface relevant articles and explore

treatment options to reduce the unwanted variation of

care and give time back to their patients.

The process Perform Clinical Assessment for On-

cology treatment (Figure 5) starts the cognitive doctor

advisor, which helps a doctor (i.e., the user) to ana-

lyze patient’s medical records and previous exams by

highlighting the potentially signiﬁcant aspects for a

given patient disease. The advisor monitors the user’s

interactions to learn and provide recommendations.

The activity Analyze Patient’s Medical Records

is detailed following the CogBPMN’s Recommenda-

tion subprocess type (Figure 1) through the subpro-

cess presented in Figure 7. The advisor analyses the

aspects visualized by the user about the patient cur-

rent medical status (activity Analyze Patient Current

Medical Status). The advisor learns from this (activity

Analyze Medical Knowledge), and provides recom-

mendations for future actions information. The user

makes decisions about the provided recommendations

(activity Decide about Relevant Medical Recommen-

dations).

The next cognitive activity of the oncology pro-

cess, Analyze Analogue Cases (Figure 5), is detailed

https://www.ibm.com/products/clinical-decision-

support-oncology

CogBPMN: Representing Human-computer Symbiosis in the Cognitive Era

855

Figure 5: Perform Clinical Assessment for Oncology treatment process - main process.

following the CogBPMN’s Recommendation with

feedback and evidence subprocess (Figure 4) through

the subprocess presented in Figure 6. The advisor an-

alyzes characteristics of other patients’ analog cases

by taking into account: (i) the feedback about recom-

mendations provided to such analog cases; (ii) the evi-

dence used to support such recommendations and the

feedback they received; and, (iii) the patient current

medical record (activity Analyze Analogue Cases).

Afterward, in the activity Analyze Related Domain

Data Sources, the advisor uses the r

esum

e about ana-

log cases generated in the previous activity to pro-

vide recommendations about the current patient case

and the evidence, such as books, papers or medical

records of other patients, to support the recommen-

dations. Then, the user provides feedback about the

recommendations (activity Provide Feedback about

Recommendation) and the evidence (activity Provide

Feedback about Evidence). Back to the main process

(Figure 5), the doctor decides if the patient should do

complementary exams.

The next oncology process’ cognitive activity

Identify Potential Treatment Options (Figure 5) is de-

tailed following the CogBPMN’s Recommendation

with evidence subprocess (Figure 2) through the sub-

process presented in Figure 8. The advisor analyzes

the case information and identiﬁes a prioritized list of

treatment options by associating then with a set of ev-

idence that supports such a list (activity Generating

Treatment Options). The advisor monitors the treat-

ment chosen by the doctor and uses such informa-

tion when making new recommendations about treat-

ments. Then, the doctor explores the recommenda-

tions and evidence (activity Explore Treatment Op-

tions and Evidence).

Finally, the doctor discusses potential treatments

with the patient (activity Discuss Potential Treatment

Options), and, if a decision about the treatment rises,

the doctor updates the patient’s medical record (activ-

ity Update Current Medical Record).

5 FINAL REMARKS AND

FUTURE WORK

Several works have evaluated the cognitive load

to understand process models in speciﬁc BPM no-

tations (Gruhn and Laue, 2006)(Gruhn and Laue,

2009)(Holschke et al., 2009). More recently, some

authors have presented the impact of cognitive sys-

tems in BPM systems. Two major theoretical contri-

butions on that subject are (Rich, 1979): 1. a frame-

work of how cognitive systems will impact BPM; and,

2. a meta-model called Plan-Act-Learn for Cogni-

tive-Enabled processes.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

856

Figure 6: Recommendation with feedback and evidence subprocess of doctor advisor process.

Figure 7: Recommendation subprocess of doctor advisor

process.

Figure 8: Recommendation with evidence subprocess of

doctor advisor process.

However, to the best of our knowledge, no work pro-

posed the extension of BPMN or other BPM notation

to model cognitive related activities

, which is the

Research conducted using Google Scholar (http://

scholar.google.com). It consisted of ﬁnding extensions for

goal of this paper. Therefore, this paper introduces an

innovative approach to model cognitive activities as

BPMN subprocesses. The proposed approach encom-

passes four complementary recommendation subpro-

cesses. The use of the approach to model the Doctor

Oncology scenario makes clear those subprocesses

should be used to model different kinds of recommen-

dations.

Our proposal points out which parts of a process

could be modeled as a cognitive subprocess along

with which kinds of cognitive activities should be

modeled if the process includes recommendations,

evidence, or feedback. The proposal is in the abstract

conceptual level presenting some directions on what

should be considered by the subprocesses’ activities.

Hence it does not present how advisors should be

combined nor which algorithmic approaches should

be used to keep the knowledge base evolving, which

is leveraged to the implementation phase.

We intend to extend CogBPMN to model other

kinds of cognitive subprocesses, such as the learn-

ing process itself. In this work, the learning process

is presented as an intrinsic characteristic of the rec-

ommendations subprocesses. However, the modeler

should be able to represent which classes of learn-

ing algorithms (or models) that are more suitable for

each type of subprocesses. Some goals are better or

only achievable with particular algorithms. For ex-

ample, if the subprocess requires that evidence should

support the recommendations, neural networks

can-

not be used (Adomavicius and Tuzhilin, 2005). The

proposed subprocesses learn in the sense that previ-

Business Process Models; several were found: for SOA

(Service-Oriented Architecture), aspects, and model-driven

BPM, among others.

Weights in Artiﬁcial Neural Networks are not easily in-

terpretable.

CogBPMN: Representing Human-computer Symbiosis in the Cognitive Era

857

ously acquired knowledge are (or should be) lever-

aged by models in their predictions. Learning could

be achieved by retraining the models.

Moreover, we are in the process of deﬁning visual

elements in BPMN to represent the four recommenda-

tions cognitive subprocesses. For example, we could

use a speciﬁc icon (like the one depicted in Figure 9)

for an activity expanded by a cognitive subprocess.

Figure 9: Icon for CogBPMN activity.

In the following, we intend to evaluate the proposed

notation by surveying experts in BPM and BPMN,

and domain experts to discuss real business processes

exploring opportunities to take advantage of cogni-

tive technology to empower human-computer inter-

action. We also intend to evaluate the proposal in

modeling other kinds of processes like transactional

and knowledge-intensive processes in the oil and gas

domain. As another future work, we plan to evolve

CogBPMN with other cognitive subprocesses types

than the proposed four in the current approach, e.g.,

conversation subprocess where the advisor identiﬁes

the willingness of the user by understanding the user’s

questions through tracking questions and the provided

answers and the feedback of the user to which ques-

tions were correctly answered.

REFERENCES

Adomavicius, G. and Tuzhilin, A. (2005). Toward the

next generation of recommender systems: A sur-

vey of the state-of-the-art and possible extensions.

IEEE transactions on knowledge and data engineer-

ing, 17(6):734–749.

Alotaibi, Y. and Liu, F. (2017). Survey of business process

management: challenges and solutions. Enterprise In-

formation Systems, 11(8):1119–1153.

Chinosi, M. and Trombetta, A. (2012). BPMN: An Intro-

duction to the Standard. Computer Standards & Inter-

faces, 34(1):124–134.

Di Ciccio, C., Marrella, A., and Russo, A. (2015).

Knowledge-intensive processes: characteristics, re-

quirements and analysis of contemporary approaches.

Journal on Data Semantics, 4(1):29–57.

Grudin, J. (2009). Ai and hci: Two ﬁelds divided by a com-

mon focus. Ai Magazine, 30(4):48–48.

Gruhn, V. and Laue, R. (2006). Complexity metrics for

business process models. In International Conference

on Business Information Systems (BIS), pages 1–12.

Citeseer.

Gruhn, V. and Laue, R. (2009). Reducing the cognitive

complexity of business process models. In 2009 8th

IEEE International Conference on Cognitive Infor-

matics, pages 339–345. IEEE.

Holbrook, J. (2017). Human-centered machine learning.

https://medium.com/google-design/human-centered-

machine-learning-a770d10562cd. Accessed in

February 14th, 2020.

Holschke, O., Rake, J., and Levina, O. (2009). Granular-

ity as a cognitive factor in the effectiveness of busi-

ness process model reuse. In International Confer-

ence on Business Process Management, pages 245–

260. Springer.

Hull, R. and Nezhad, H. R. M. (2016). Rethinking BPM in

a Cognitive World: Transforming how we learn and

perform business processes. In International Confer-

ence on Business Process Management, pages 3–19.

Springer.

Kelly, J. E. (2015). Computing, cognition and the future of

knowing. Whitepaper, IBM Research, 2.

Kelly III, J. E. and Hamm, S. (2013). Smart machines:

IBM’s Watson and the era of cognitive computing.

Columbia University Press.

Kidd, A. (2012). Knowledge acquisition for expert systems:

A practical handbook. Springer Science & Business

Media.

KPMG (2015). Robotic revolution - separating hype

from reality. https://home.kpmg/cn/en/home/insights/

2015/10/robotic-revolution-o-201510.html. Accessed

in February 14th, 2020.

Mitchell, T. M. and et al (1997). Machine learning. Mc-

Graw Hill, 45(37):870–877.

Modha, D. S., Ananthanarayanan, R., Esser, S. K., Ndi-

rango, A., Sherbondy, A. J., and Singh, R. (2011).

Cognitive computing. Communications of the ACM,

54(8):62–71.

Netto, J. M., Franc¸a, J. B., Bai

ao, F. A., and Santoro,

F. M. (2013). A notation for knowledge-intensive pro-

cesses. In Proceedings of the 2013 IEEE 17th Interna-

tional Conference on Computer Supported Coopera-

tive Work in Design (CSCWD), pages 190–195. IEEE.

Nezhad, H. R. M. and Akkiraju, R. (2014). Towards cog-

nitive bpm as the next generation bpm platform for

analytics-driven business processes. In International

Conference on Business Process Management, pages

158–164. Springer.

OMG (2011). Notation (BPMN) version 2.0. OMG Speci-

ﬁcation, Object Management Group.

Rich, E. (1979). User modeling via stereotypes. Cognitive

science, 3(4):329–354.

Turski, W. M. and Maibaum, T. S. (1987). Speciﬁcation of

computer programs. Addison-Wesley Longman Pub-

lishing Co., Inc.

Van der Aalst, W. M. (2013). Business process manage-

ment: a comprehensive survey. ISRN Software Engi-

neering, 2013.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

858