A Conceptual Model of Actors and Interactions

for the Knowledge Discovery Process

Lauri Tuovinen

Biomimetics and Intelligent Systems Group, University of Oulu,

P.O. Box 4500, FI-90014 Oulu, Finland

Keywords:

Knowledge Discovery in Data, Process Model, Computer-supported Collaboration, Intelligent Systems.

Abstract:

The knowledge discovery process is traditionally viewed as a sequence of operations to be applied to data; the

human aspect of the process is seldom taken into account, and when it is, it is mainly the roles and actions

of domain and technology experts that are considered. However, non-experts can also play an important role

in knowledge discovery, and furthermore, the role of technology in the process may also be substantially

expanded from what it traditionally has been, with special software facilitating interactions among human

actors and even operating as an actor in its own right. This diversiﬁcation of the knowledge discovery process

is helpful in ﬁnding tenable solutions to the new problems presented by the current deluge of digital data, but

only if theprocess model used to manage the process adequately represents the variety of forms that the process

can take. The paper addresses this requirement by presenting a conceptual model that can be used to describe

different types of knowledge discovery processes in terms of the actors involved and the interactions they have

with one another. Additionally, the paper discusses how the interactions can be facilitated to provide effective

support for each different type of process. As a future perspective, the paper considers the implications of

intelligent software taking on responsibilities traditionally reserved for human actors.

1 INTRODUCTION

Twenty years ago, the seminal process model for

knowledge discovery in data (KDD) was presented

(Fayyad et al., 1996). In this model, the process is di-

vided into ﬁve steps: selection, preprocessing, trans-

formation, data mining, and interpretation/evaluation.

Today, there are a number of KDD process models of-

fering various augmentations and alternative perspec-

tives to the seminal one, but the big picture has not

changed radically, with the sequence of steps laid out

by Fayyad et al. still being found at the heart of many

of the more recent proposals. In fact, it is arguably

the case that even if it is not explicitly acknowledged,

this model is present in every KDD effort, since it pre-

scribes a set of operations that are always required to

get from raw data to actionable knowledge, even if

they do not generally occur in a neat waterfall-style

sequence.

The problem with the established KDD process

model is not that it does a poor job of representing

what it is intended to represent, but rather that what

it represents is not the whole picture of KDD but just

one aspect of it. The human aspect - the actors who

participate in the KDD process and the interactions

among them - is not addressed at all, and although

there are some models that do account for it in some

way, they are increasingly unsuitable for represent-

ing the variety of human-human and human-machine

interactions that may be required in order to success-

fully complete a KDD effort. There is, therefore, a

need for a collaborative process model that covers the

full spectrum of actors and interactions involved.

A truly collaborativemodel of the KDD process is

an important goal because traditional process models

and methodologies are not designed to adequately re-

spond to the challenges posed by the data deluge we

are currently experiencing. According to a report pub-

lished in 2012, 1.8 zettabytes (1.8∗ 10

bytes) of data

were generated globally in 2011 (Federal Big Data

Commission, 2012), and the rate is likely to be even

higher today. Even the volume of data available on

the Web to any Internet-connected individual is over-

whelming, and it is relatively effortless and inexpen-

sive to collect more for one’s own purposes using var-

ious sensors, such as those we already carry with us

everywhere in our smartphones. The question, there-

fore, is not how we can obtain data, but how we can

beneﬁt from it.

Traditionally, the interactions that drive the KDD

240

Tuovinen, L.

A Conceptual Model of Actors and Interactions for the Knowledge Discovery Process.

DOI: 10.5220/0006045902400248

In Proceedings of the 8th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2016) - Volume 1: KDIR, pages 240-248

ISBN: 978-989-758-203-5

process take place between domain experts and tech-

nology experts; these two actor types collaborate to

develop an understanding of what data is available

and what knowledge could be gained from it, deﬁne

a problem, design a solution and evaluate the results.

Sometimes, however, the experts rely on non-expert

actors for some crucial resource, which makes the in-

teractions more complicated. Furthermore, there is

now a new trend where the KDD process is driven by

the needs of a non-expert actor, effectively reversing

the traditional roles of experts and non-experts. The

process model of KDD needs to be able to represent

the dynamics of all these different collaborative rela-

tionships that the practice of KDD may entail.

This paper presents a conceptual model to serve as

the basis of understanding the actors of the KDD pro-

cess and the collaborative interactions among them.

Besides human actors, the model includes technol-

ogy as a special type of actor; traditionally, the role

of technology in KDD has been to serve as a tool to

be applied by technology experts, but advances in the

ﬁeld of artiﬁcial intelligence are making it increas-

ingly feasible for autonomous software to carry out

KDD tasks that have previously required a human ac-

tor. As the data deluge continues to multiply in vol-

ume, such software will be an essential part of any

effort to make sense of it, and therefore its role in the

KDD process must be deﬁned by the process model

just as the roles of the human actors are.

The remainder of this paper is organized as fol-

lows: Section 2 reviews literature on KDD process

models and other relevant topics, showing how the

new work presented in the paper advances beyond

the current state of the art. Sections 3 through 5

present the main components of the proposed con-

ceptual model, with Section 3 focusing on the actors,

Section 4 on the interactions, and Section 5 on dif-

ferent ways of facilitating the interactions. Section 6

examines the relative strenghts and weaknesses of the

proposed model and other KDD process models, and

suggests a way of applying the model in practice. Sec-

tion 7 discusses some noteworthy future perspectives,

and Section 8 concludes the paper.

2 BACKGROUND

Following the publication of the KDD process model

of (Fayyad et al., 1996), a number of derivative mod-

els were developed, as described in the survey of

(Mariscal et al., 2010). In 2000, the CRISP-DM

model (Wirth and Hipp, 2000) was published, in-

tended as a standard model for the KDD process

and incorporating elements or inﬂuences from sev-

eral of the previously proposed models. This was

followed by another set of derivative models, as well

as another standardization effort, although ultimately

it appears that the development of CRISP-DM 2.0

has ceased without ever yielding any concrete output

(McCormick, 2007).

The idea of identifying the actors of the KDD pro-

cess, their roles in the process and their relationships

with one another can be found already in some early

work such as the model proposed in (Brachman and

Anand, 1996). The idea carried over to the Internet-

enabled model of (B¨uchner et al., 1999) – a derivative

designed speciﬁcally for knowledge discovery from

Web data – and also to CRISP-DM, which acknowl-

edges the importance of understanding the problem at

hand from the business perspective of the customer

as a prerequisite for examining it from the technical

perspective of the KDD analyst.

Among relevant work published after CRISP-DM,

particularly interesting are the RAMSYS methodol-

ogy for remote collaborative KDD efforts (Moyle and

Jorge, 2001), the knowledge exchange perspective of

(Diamantini et al., 2006), and the knowledge fusion

model of (Horeis and Sick, 2007). The ASUM-DM

methodology (Haffar, 2015) of IBM is also relevant,

being an augmented version of CRISP-DM. However,

the ﬁrst process model that properly acknowledges

the role of non-expert actors in the KDD process was

proposed by the author (Tuovinen, 2014).

The author’s model builds upon previous research

on issues related to the involvement of non-experts

in KDD; for instance, there was already a signiﬁ-

cant body of work on dealing with the potential ethi-

cal problems arising from the use of personal data in

KDD. However, such issues were previously always

considered in isolation, as technical problems rather

than something to be accounted for on the level of the

KDD process. The author’s model, in contrast, adopts

the perspective that non-experts should be treated by

the process model as actors with important contribu-

tions that the process should facilitate, and as stake-

holders with legitimate expectations that the process

should aim to satisfy.

In this paper, the author’s previously published

work on modeling the KDD process in terms of actors

and interactions is extended in the following ways:

• In addition to the four actor types inherited from

previous work, the paper deﬁnes four interaction

types that, together with the actor types, can be

used to construct different types of KDD pro-

cesses.

• Using these basic elements, the paper presents a

more detailed and formal account of three previ-

ously identiﬁed KDD process types, and addition-

A Conceptual Model of Actors and Interactions for the Knowledge Discovery Process

241

ally of one type not covered by previous work,

where the process is driven by a non-expert actor.

• Having identiﬁed and described the four different

process types, the paper discusses possible ways

of supporting them by using technology to facili-

tate key interactions.

These extensions constitute the novel scientiﬁc con-

tribution of the paper.

3 TYPES OF ACTORS

Following the lead of previous research, we begin by

identifying three principal types of human actors in

the KDD process: domain experts, technology ex-

perts and non-experts. Additionally, interaction with

computing technology is an essential part of any KDD

process, so we shall consider that the fourth type of

actor. We can now visually represent the relation-

ships between these different actor categories using

a triangle, with the human actors at the vertices and

technology at the center, as shown in Figure 1.

The roles played by the actors vary depending on

which speciﬁc type of KDD process we are examin-

ing. Different types of processes can be character-

ized based on which of the three human actor types

are mainly involved and which of them interact with

one another; thus we can visualize KDD process types

by highlighting different sections of the triangle as

in Figure 2. Here we identify four different process

types, which we shall now look at in more detail.

The standard version of the KDD process is the

one where only the base of the triangle, representing

Figure 1: The four main types of actors in the KDD process.

Technology is both a mediator of interactions among the

three types of human actors (vertices of the triangle) and an

actor in its own right.

the collaborativerelationship between domain experts

and technology experts, is involved(Figure 2(a)). The

relationship, as already outlined in the introduction,

consists of interactions for specifying objectives, de-

veloping a solution and evaluating the outcome; in-

teractions with technology are required to execute

the necessary computations and examine their results.

This process type is already quite well covered by ex-

isting process models such as CRISP-DM.

Involving the third vertex of the triangle, repre-

senting non-expert human actors, in different ways

yields different variations or special cases of the stan-

dard KDD process. Assuming that the process is still

driven by the expert actors, there are two ways in

which the non-experts can be involved: they can be

a source of data that will be analyzed in the KDD ef-

fort, or they can provide computing resources to be

used in the effort for data analysis. In the former case,

their main interaction is with the domain experts, with

whom they agree on the terms under which their data

may be used (Figure 2(b)); in the latter case, a form of

KDD known as volunteer computing, the main inter-

action is with the technologyexperts, who take care of

the logistics of recruiting volunteers and distributing

tasks (Figure 2(c)).

In the volunteer computing case, the technology

actor has a role that differs somewhat from its role

in the standard KDD process. The data processing in

this case is done on the computers of the volunteers,

and to coordinate this sub-process, there needs to be a

system that divides the data to be analyzed into parti-

tions, sends them out to be processed and collates the

results. Thus, in addition to its usual role, technology

acts here in a mediating role between the expert and

non-expert human actors. Perhaps the most famous

example of volunteer computing is the SETI@home

project (Korpela et al., 2015), but there are many oth-

ers; in a variation, it is the volunteers themselves and

not their computers who perform the analysis, e.g. by

classifying galaxy images (Lintott et al., 2008).

In both the above cases, it is important to ensure

that the rights and legitimate expectations of the non-

expert actors are respected. Unlike the experts, the

non-experts may have no immediate interest in the

outcome of the KDD effort, but they do typically have

interests of their own at stake; the right to privacy,

in particular, may be threatened when KDD is done

on personal data, and there is a substantial body of

research on how violating it can be avoided without

compromising the objectives of the KDD effort (Shah

and Gulati, 2015; Terzi et al., 2015). Volunteer com-

puting has not been studied as much from this per-

spective, but it also has some potential ethical issues

to be addressed (Tuovinen and R¨oning, 2009).

KDIR 2016 - 8th International Conference on Knowledge Discovery and Information Retrieval

242

(a) Standard KDD pro-

cess

(b) KDD using personal

data

computing

(d) KDD driven by a

non-expert actor

Figure 2: KDD process types illustrated. A shaded triangle section signiﬁes that the corresponding actors are involved in the

process and interact with one another. Refer to Figure 1 for a mapping of triangle vertices to actors.

It is also possible for non-expert actors to take

charge of analyzing their own data (Figure 2(d)); this

is happening, for example, with the quantiﬁed self

movement, where some knowledgeable individuals

have begun to use various wearable sensor devices

and other data sources as means to improve their per-

sonal health and well-being (Swan, 2013). In this

case, the non-expert relies on the experts’ contribu-

tions rather than the other way around, and the in-

teractions among them may be more indirect, with

the non-expert e.g. consulting sources written by the

experts instead of collaborating with them directly.

Also, the technology actor needs to partially assume

the role of technology expert by assisting the non-

expert in the construction of KDD workﬂows and the

evaluation of results.

It is clear from looking at these different varia-

tions of the KDD process that introducing the cat-

egory of non-expert human actors creates a consid-

erable amount of diversity in the roles and relation-

ships of all the actors. Since established KDD process

models do not account for the non-expert category,

they cannot adequately represent all this diversity, and

therefore cannot fully support the planning and exe-

cution of KDD processes that differ from the standard

one. This point becomes even clearer in the next sec-

tion, which views the KDD process as a sequence of

interactions and compares how the sequence plays out

in different process types.

4 TYPES OF INTERACTIONS

In order to be able to build a conceptual model of the

KDD process in terms of interactions among the ac-

tors identiﬁed in the previous section, we ﬁrst need to

deﬁne a set of interaction types that we can use to rep-

resent the dynamics of the process. Based on the in-

teractions that take place between domain experts and

technology experts in the standard version of the pro-

cess, we can identify the following four basic types:

• Negotiation: the actors establish the terms of their

collaboration, agreeing on what each partner will

contribute and on what conditions.

• Analysis: the actors establish a shared understand-

ing of a domain of mutual interest to enable them

to deﬁne a further course of action

• Assignment: the actors agree on a speciﬁc task

to be carried out by some of them to achieve an

established objective

• Delivery: the actors who carried out a task hand

over the result to those expecting them and help

them understand it.

Besides interactions between human actors, as-

signment and delivery can be used to represent the

interactions of technology experts with technology as

they set up and execute computations and examine

the results. Thus, using these four interaction types,

we can visualize the standard KDD process using se-

quence diagram notation, as shown in Figure 3.

The different variations of the standard process

discussed in Section 3 can similarly be depicted as

sequences of negotiation, analysis, assignment and

delivery interactions. Figure 4 shows extracts from

the sequence diagrams for the variations, focusing on

those parts of the process where they differ notably

from the standard case. In the case where non-expert

actors are employed as a source of personal data, the

most signiﬁcant difference is that once the experts

have completed the analysis interaction and thus es-

tablished their data requirements, interactions with

the non-experts are required to obtain the data (Fig-

ure 4(a)). Once the experts have gained access to the

data, the process can continue in the standard manner.

In the volunteer computing case, where the non-

expert actors are employed as providers of computing

resources, the main difference is in how the computa-

tions are done (Figure 4(b)). Whereas in the standard

version of the KDD process the technology experts

interact with technology for this purpose, in volunteer

computing both the experts and the non-experts inter-

act with technology, and additionally there is a large

A Conceptual Model of Actors and Interactions for the Knowledge Discovery Process

243

Figure 3: Depiction of the standard version of the KDD process as a sequence of interactions among domain experts, technol-

ogy experts and technology.

number of interactions between individualtechnology

actors - the computing server that distributes the jobs

and the clients that process them. Not shown in the

ﬁgure are the community interactions characteristic of

volunteer computing efforts, where, for instance, the

experts share information on the progress of the effort

with the volunteers.

Finally, in the case where the KDD process is

driven by a non-expert actor, the entire process differs

considerably from the standard version, with the role

of technology substantially expanded and the roles of

human experts diminished accordingly (Figure 4(c)).

To compensate for the non-expert’s lack of essen-

tial domain knowledge and technical skills, it is es-

pecially important in this case for the interactions to

be adequately supported, so that the non-expert can

obtain the required knowledge and apply it to set up

a KDD workﬂow. However, also in the other three

cases supporting the process is largely a matter of sup-

porting the interactions.

Supporting the various types of interactions that

the KDD process in its various incarnations may in-

volve is a vast, complex and multidisciplinary issue

that cannot be addressed here in much detail. How-

ever, if we limit our discussion to a technological per-

spective, we can discern two principal types of sup-

port that computing technology can provide: we can

use computers to facilitate interactions between hu-

man actors, or the computers themselves can interact

with humans (or other computers) in more advanced

ways. The next section takes a closer look at these

possibilities.

5 FACILITATING THE

INTERACTIONS

The most obvious way in which technology can fa-

cilitate collaborative interactions among the actors of

the KDD process is by providing them with a commu-

nication channel. A wide range of options is already

available for this purpose, from basic text-based chan-

nels such as email and instant messaging services to

more advanced teleconferencing solutions with fea-

tures such as live audio/video and screen sharing.

However, while undoubtedly useful, well-established

and generic technologies such as these are not partic-

ularly interesting in this context; we are more inter-

ested in facilitating communication in the KDD pro-

cess speciﬁcally, and especially in supporting inter-

actions involving non-expert actors, which are not as

well understood as the interactions of the standard

process.

One possible bottleneck in interactions involving

non-experts is setting up the interaction: before col-

laboration can take place, the collaborators need to

be introduced to one another somehow. Pioneering

work addressing this problem has been done in the

area of volunteer computing, where there are frame-

works such as BOINC (Anderson, 2004) that make

it simpler both for experts to set up projects and for

non-experts to contribute to them. Various crowd-

sourcing platforms in general are relevant here in that

they establish a convenient way for people in need of

a service to ﬁnd and negotiate with people who are

able and willing to provide the service, be it data pro-

cessing or, for example, transport as in the case of

Uber. A similar approach would conceivably work in

other forms of KDD as well, with a special software

platform bringing the actors together and providing

them with easy-to-use tools for specifying what they

require and what they offer in exchange for it.

In terms of the four interaction types deﬁned in the

previous section, the platform outlined above would

facilitate negotiation, assignment and delivery inter-

actions among human actors, but not analysis inter-

actions. In most versions of the KDD process, anal-

ysis is where the domain and technology experts in-

teract to establish a shared understanding of the prob-

lem to be solved and the data and methods available

for tackling it. This interaction cannot be modeled as

an exchange of service and compensation; it requires

closer collaboration among the actors, and therefore

KDIR 2016 - 8th International Conference on Knowledge Discovery and Information Retrieval

244

(a) KDD using personal data (b) KDD using volunteer computing

Figure 4: Variations of the standard interaction sequence of the KDD process.

facilitating it requires a different approach, one that

acknowledges that the result of the analysis is con-

structed not by one actor for the beneﬁt of another but

by both actors together.

Again, there are general-purposesolutions for col-

laborative work that are technically relevant but not

particularly interesting; for instance, cloud services

such as Google Docs and Ofﬁce Online, which en-

able real-time collaborative editing of shared docu-

ments, could be used to develop artifacts represent-

ing the outcome of the analysis. In fact, collaborative

editing is conceivably a viable paradigm for the fa-

cilitation of analysis interactions in the KDD process,

but the format of the artifacts and the tools provided

by the editor would have to be much more specialized

than those found in generic ofﬁce software suites. For

example, by enabling the actors to specify available

datasets, represent their internal semantics and iden-

tify potentially important relationships among them,

the editor could help the actors understand what they

could accomplish using the available data and which

datasets they would need to combine to solve a given

KDD problem.

So far, we have only considered the facilitation

of interactions between human actors. KDD hav-

ing been traditionally subsumed under computer sci-

ence, it is probably true that interactions with tech-

nology are currently better understood than interac-

tions between humans, which require a more multi-

disciplinary approach. Nevertheless, interacting with

technology is far from trivial, and the importance of

supporting these interactions effectively will increase

as the data deluge gets more and more difﬁcult to con-

trol, so we should take here some time to look at how

they could be facilitated better.

In the standard version of the KDD process, hu-

man actors interact with technology via assignment

and delivery interactions where the role of the tech-

nology actor is limited to executing the computations

speciﬁed by the human actor and returning the results.

Both the assignment and the delivery can be made

more collaborative to improve this sub-process. To

facilitate assignment, the speciﬁcation of KDD work-

ﬂows can be partially automated; there is already a

substantial body of interesting research in this area, as

demonstrated by a 2013 survey of KDD software sys-

tems capable of providing at least some level of intel-

ligent assistance to the user (Serban et al., 2013). As

one would expect, the ideal intelligent assistant does

not yet exist: the systems reviewed are generally lim-

ited in some way, and those that are not require the

user to be experienced enough to conﬁgure the algo-

rithms and compose the workﬂow without assistance.

The paper notes the limitations and proceeds to offer

some future directions toward more advanced intel-

ligent assistants, including the observation that they

should be based on a collaborative model of KDD.

To facilitate delivery, new interactivevisualization

techniques can be developed to make it easier for the

human actor to grasp the signiﬁcance of the results.

The need for closer integration of information visual-

ization with KDD has given rise to the research ﬁeld

known as visual analytics. Several state-of-the-art

surveys of this ﬁeld have been carried out in recent

A Conceptual Model of Actors and Interactions for the Knowledge Discovery Process

245

years; academic research is reviewed in (Sun et al.,

2013) and commercial software systems in (Zhang

et al., 2012). (Holzinger, 2013) also reviews a consid-

erable amount of relevant work, advocating the inte-

gration of KDD with human-computer interaction in

general. An interesting example of a technology that

may prove useful for this purpose in the future is pro-

vided by (Donalek et al., 2014), where the possibility

of using virtual reality platforms for information vi-

sualization is explored.

In summary, there are two principal ways in which

new technology can help facilitate interactions in the

KDD process:

• Interactions among human actors can be facili-

tated by matching potential collaborators and pro-

viding them with collaboration tools designed

speciﬁcally for KDD tasks.

• Interactions between human actors and technol-

ogy can be facilitated by developing intelligent

software capable of taking on a more active role

in the execution of KDD processes.

Although the role of technology in the KDD process

would be somewhat expanded by these developments,

negotiation and analysis interactions would still take

place exclusively between human actors, with tech-

nology acting only as a facilitator. Section 7 discusses

future developments that would transform the process

in a more fundamental manner.

6 NOVELTY AND UTILITY

The main advantage of the proposedmodel over exist-

ing KDD process models is that it presents a broader

view of the KDD process by including the category

of non-expert actors. As a consequence, the proposed

model is more suitable for representing process ﬂow

in variants of the KDD process where non-experts

play a substantial part. The “quantiﬁed self” variant,

in particular, is difﬁcult to reconcile with established

models, which are built on the assumption that the

process is driven by expert actors.

Having a broader perspective on who the partici-

pants of the KDD process are also results in a broader

perspective on what the objectives of the process are.

In the traditional view of KDD, the purpose of the

process is to extract value from data for the owner of

the data, but this is just one aspect of the big picture,

representing the point of view of one participant; the

more participants there are, the more purposes there

are, and some of the purposes may conﬂict with one

another. In such cases it is essential to steer the pro-

cess such that the nature of the conﬂict is identiﬁed

and measures to resolve it are taken.

In its present state, the proposed model is mainly

a conceptual one, which limits its utility as a practical

tool for the management of KDD efforts. In compar-

ison with, for example, CRISP-DM, the model lacks

an overarching task structure and detailed guidance

for the completion of each task. Some amount of

structure and guidance is provided by the classiﬁca-

tion of interactions and the description of KDD pro-

cess types as interaction sequences, but there is a con-

siderable amount of practical veriﬁcation and reﬁne-

ment to be done here.

In fact, probably the most effective way to apply

the proposed model in its current form is to adopt

the task structure of CRISP-DM – or some other es-

tablished KDD process model – and use the inter-

action model to complement it. For example, the

top-level task structure of CRISP-DM consists of six

phases: business understanding, data understanding,

data preparation, modeling, evaluation and deploy-

ment. If we substitute “business understanding” with

“domain understanding”, the phases are completely

generic and not in any way dependent on who the pro-

cess actors are.

We can thus conceive a meta-process for combin-

ing the two models, where the ﬁrst step is to identify

the actors taking part in the KDD process. It is then

possible to map each phase of the standard process

model to the interactions among the actors that take

place in each phase. Now, for guidance on how to ex-

ecute the process, we can refer to both the documen-

tation of the standard model and the ideas discussed

here on how the interactions can be facilitated.

7 DISCUSSION

In KDD efforts driven by non-expert actors, it is nec-

essary for the technology actor to assume a more ac-

tive role in analysis interactions than it does in more

traditional versions of the process. Currently there is

software available that can do this in a limited manner,

using preconﬁgured algorithms to extract knowledge

from predeﬁned data sources (e.g. a speciﬁc type of

activity monitor). If the non-expert wishes to tran-

scend such limitations, they need to learn to use a

more generic KDD software suite, effectively becom-

ing a technology expert as a result.

For us to be able to take a more decisive step to-

ward this new type of KDD process, the technology

actor needs to become more intelligent so that it can,

in effect, play the role of both domain expert and tech-

nology expert and thus assist the non-expert human

actor with the task of turning their requirements into

KDIR 2016 - 8th International Conference on Knowledge Discovery and Information Retrieval

246

a speciﬁcation for a KDD solution. The construction

of the solution and the interpretation of the results –

human-technology interactions in expert-drivenKDD

processes – would be handled autonomously by the

technology actor. The role of the human actor in this

case would be to collect the data, specify what they

want to discover from it, and, having received the re-

sults, decide on next steps.

Clearly the ability of the technology actor to con-

tribute to analysis interactions as an active participant

would also substantially change the nature of KDD

processes driven by expert human actors. With the

technology actor taking over some of the responsibil-

ities of the human actors, the latter could spend more

of their time on higher-level analysis and planning ac-

tivities. They would thus be able to address KDD

problems on a higher level of abstraction, allowing

them to tackle more complex problems and making

the data deluge more manageable.

Taking yet a step further, we can consider the

possibility of collaboration among technology actors.

This would involve autonomously operating units of

KDD software interacting with one another by re-

questing and providing services; thus, for example,

a technology actor engaging in an analysis interaction

with a human actor could enlist the help of another

technology actor capable of offering special exper-

tise relevant to the problem at hand. To a certain ex-

tent, this is already happening in various schemes for

agent-based KDD such as those mentioned in (Klusch

et al., 2003), but so far, the role of the agents has been

subject to the same limitations as the role of technol-

ogy in the KDD process in general, the agent-based

approach mainly serving as a strategy for distributing

computations.

In the future, we can expect KDD agents to even

have the ability to engage in negotiation-type interac-

tions with one another and with human actors. Such

agents would serve as autonomous proxies for their

owners, negotiating and performing the exchange of

services and compensation without requiring human

intervention as long as the terms offered are within

the limits of their authorization. The beneﬁts of agent-

based KDD, particularly those pertaining to distribu-

tion of computational operations, would thus be ex-

panded to cover a wider range of KDD tasks, and the

range of options available for managing the data del-

uge would be similarly expanded.

In terms of the triangle diagrams of Section 3, the

logical next step is a KDD process where all the ver-

tices of the triangle are equally active, with each indi-

vidual KDD effort having the optimal combination of

actors as determined by the requirements of the prob-

lem at hand and the availability of resources. With a

sufﬁciently powerful remote collaboration platform,

geographical distribution of the actors is not a prob-

lem; what may become one, however, is coordination

of the interactions among the actors, as there is no up-

per limit to the complexity of the KDD processes that

can be constructed in this way. Designing a process

model capable of coping with this complexity is nec-

essary for us to be able to cope with the increasing

complexity of KDD problems.

8 CONCLUSION

Digital data is potentially a highly valuable resource

for a wide range of users, from individuals to large

organizations. However, realizing this potential de-

pends on ﬁnding the most appropriate human and

technological resources for the task and using them

effectively. This, in turn, involves an intricate web

of collaborative interactions that the established KDD

process model is not very well equipped to represent

or support. The model particularly fails to account for

the expanding roles of computers and non-expert hu-

mans in the process, both of which are necessitated

by the new challenges arising from the big data phe-

nomenon.

In this paper we presented a conceptual model in-

tended to pave the way for a more detailed KDD pro-

cess model that represents the collaborative aspect of

the process better than the established model. The

conceptual model describes the basic actor and inter-

action types that occur in KDD and shows that sev-

eral different KDD process types can be derived from

these. Practitioners of KDD can use the model to

identify the actors involved in a given KDD effort,

the role of each actor and the nature of the interac-

tions between each pair of actors; this knowledge is

required for the practitioner to be able to support the

process, which is largely a matter of facilitating the

interactions effectively.

ACKNOWLEDGEMENTS

The work presented here extends the author’s disser-

tation; the author would like to thank Prof. Alan F.

Smeaton for discussions that inspired this follow-up.

REFERENCES

Anderson, D. P. (2004). BOINC: A system for public-

resource computing and storage. In Proceedings of

A Conceptual Model of Actors and Interactions for the Knowledge Discovery Process

247

the Fifth IEEE/ACM International Workshop on Grid

Computing, pages 4–10.

Brachman, R. J. and Anand, T. (1996). The process of

knowledge discovery in databases. In Advances in

Knowledge Discovery and Data Mining, pages 37–57.

American Association for Artiﬁcial Intelligence.

B¨uchner, A. G., Mulvenna, M. D., Anand, S. S., and

Hughes, J. G. (1999). An internet-enabled knowledge

discovery process. In Proceedings of the 9th Interna-

tional Database Conference, pages 13–27.

Diamantini, C., Potena, D., Domenico, and Smari, W. W.

(2006). Collaborative knowledge discovery in

databases: A knowledge exchange perspective. In

Proceedings of the AAAI Fall Symposium on Semantic

Web for Collaborative Knowledge Acquisition, pages

24–31.

Donalek, C., Djorgovski, S. G., Cioc, A., Wang, A., Zhang,

J., Lawler, E., Yeh, S., Mahabal, A., Graham, M.,

Drake, A., Davidoff, S., Norris, J. S., and Longo, G.

(2014). Immersive and collaborative data visualiza-

tion using virtual reality platforms. In Proceedings of

the 2014 IEEE International Conference on Big Data,

pages 609–614.

Fayyad, U., Piatetsky-Shapiro, G., and Smyth, P. (1996).

The KDD process for extracting useful knowledge

from volumes of data. Communications of the ACM,

39(11):27–34.

Federal Big Data Commission (2012). Demystifying big

data: A practical guide to transforming the business

of government. Technical report, TechAmerica Foun-

dation.

Haffar, J. (2015). Have you seen ASUM-DM? Blog entry,

retrieved 16 Sep, 2016. https://developer.ibm.com/

predictiveanalytics/2015/10/16/have-you-seen-asum-

dm/.

Holzinger, A. (2013). Human-computer interaction and

knowledge discovery (HCI-KDD): What is the ben-

eﬁt of bringing those two ﬁelds to work together? In

Availability, Reliability, and Security in Information

Systems and HCI: Proceedings of the International

Cross-Domain Conference and Workshop, pages 319–

328.

Horeis, T. and Sick, B. (2007). Collaborative knowledge

discovery & data mining: From knowledge to experi-

ence. In Proceedings of the 2007 IEEE Symposium on

Computational Intelligence and Data Mining, pages

421–428.

Klusch, M., Lodi, S., and Moro, G. (2003). Agent-

based distributed data mining: The KDEC scheme.

In Klusch, M., Bergamaschi, S., Edwards, P., and

Petta, P., editors, Intelligent Information Agents:

The AgentLink Perspective, pages 104–122. Springer

Berlin Heidelberg.

Korpela, E. J., Siemion, A. P. V., Werthimer, D., Lebofsky,

M., Cobb, J., Croft, S., and Anderson, D. (2015). The

next phases of SETI@home. In Proceedings of SPIE

9606, Instruments, Methods, and Missions for Astro-

biology XVII.

Lintott, C. J., Schawinski, K., Slosar, A., Land, K., Bam-

ford, S., Thomas, D., Raddick, M. J., Nichol, R. C.,

Szalay, A., Andreescu, D., Murray, P., and Vanden-

berg, J. (2008). Galaxy Zoo: Morphologies derived

from visual inspection of galaxies from the Sloan Dig-

ital Sky Survey. Monthly Notices of the Royal Astro-

nomical Society, 389(3):1179–1189.

Mariscal, G., Marb´an, O., and Fern´andez, C. (2010). A sur-

vey of data mining and knowledge discovery process

models and methodologies. The Knowledge Engineer-

ing Review, 25(2):137–166.

McCormick, K. (2007). CRISP-DM 2.0. Blog entry, re-

trieved 16 Sep, 2016. http://keithmccormick.com/

crisp-dm-20/.

Moyle, S. and Jorge, A. (2001). RAMSYS - a methodology

for supporting rapid remote collaborative data mining

projects. In ECML/PKKD01 Workshop on Integrating

Aspects of Data Mining, Decision Support and Meta-

Learning, pages 20–31.

Serban, F., Vanschoren, J., Kietz, J.-U., and Bernstein, A.

(2013). A survey of intelligent assistants for data anal-

ysis. ACM Computing Surveys, 45(3):article 31.

Shah, A. and Gulati, R. (2015). Contemporary trends in

privacy preserving collaborative data mining - a sur-

vey. In Proceedings of the 2015 International Confer-

ence on Electrical, Electronics, Signals, Communica-

tion and Optimization.

Sun, G.-D., Liang, R.-H., and Liu, S.-X. (2013). A survey

of visual analytics techniques and applications: State-

of-the-art research and future challenges. Journal of

Computer Science and Technology, 28(5):852–867.

Swan, M. (2013). The quantiﬁed self: Fundamental disrup-

tion in big data science and biological discovery. Big

Data, 1(2):85–99.

Terzi, D. S., Terzi, R., and Sagiroglu, S. (2015). A survey on

security and privacy issues in big data. In Proceedings

of the 10th International Conference for Internet Tech-

nology and Secured Transactions, pages 202–207.

Tuovinen, L. (2014). From machine learning to learning

with machines: Remodeling the knowledge discovery

process. PhD thesis. University of Oulu, Finland.

Tuovinen, L. and R¨oning, J. (2009). Everybody wins: Chal-

lenges and promises of knowledge discovery through

volunteer computing. In Proceedings of the 8th Inter-

national Conference on Computer Ethics: Philosoph-

ical Enquiry, pages 821–842.

Wirth, R. and Hipp, J. (2000). CRISP-DM: Towards a stan-

dard process model for data mining. In Proceedings of

the 4th International Conference on the Practical Ap-

plications of Knowledge Discovery and Data Mining,

pages 29–39.

Zhang, L., Stoffel, A., Behrisch, M., Mittelstadt, S.,

Schreck, T., Pompl, R., Weber, S., Last, H., and Keim,

D. (2012). Visual analytics for the big data era -

a comparative review of state-of-the-art commercial

systems. In Proceedings of the 2012 IEEE Conference

on Visual Analytics Science and Technology, pages

173–182.

KDIR 2016 - 8th International Conference on Knowledge Discovery and Information Retrieval

248