=Paper=
{{Paper
|id=None
|storemode=property
|title=Building up Shared Knowledge with Logical Information Systems
|pdfUrl=https://ceur-ws.org/Vol-959/paper2.pdf
|volume=Vol-959
|dblpUrl=https://dblp.org/rec/conf/cla/DucasseFC11
}}
==Building up Shared Knowledge with Logical Information Systems==
https://ceur-ws.org/Vol-959/paper2.pdf
Building up Shared Knowledge with
Logical Information Systems
Mireille Ducassé1 , Sébastien Ferré2 , and Peggy Cellier1
1
IRISA-INSA de Rennes, France,
{ducasse, cellier}@irisa.fr
2
IRISA-University of Rennes 1, France
ferre@irisa.fr
Abstract. Logical Information Systems (LIS) are based on Logical Con-
cept Analysis, an extension of Formal Concept Analysis. This paper de-
scribes an application of LIS to support group decision. A case study
gathered a research team. The objective was to decide on a set of po-
tential conferences on which to send submissions. People individually
used Abilis, a LIS web server, to preselect a set of conferences. Start-
ing from 1041 call for papers, the individual participants preselected 63
conferences. They met and collectively used Abilis to select a shared
set of 42 target conferences. The team could then sketch a publication
planning. The case study provides evidence that LIS cover at least three
of the collaboration patterns identified by Kolfschoten, de Vreede and
Briggs. Abilis helped the team to build a more complete and relevant
set of information (Generate/Gathering pattern); to build a shared un-
derstanding of the relevant information (Clarify/Building Shared Un-
derstanding); and to quickly reduce the number of target conferences
(Reduce/Filtering pattern).
1 Introduction
Group work represents a large amount of time in professional life while many
people feel that much of that time is wasted. Lewis [13] argues that this amount
of time is even going to increase because problems are becoming more complex
and are meant to be solved in a distributed way. Each involved person has a local
and partial view of the problem, no one embraces the whole required knowledge.
Lewis also emphasizes that it is common that “groups fail to adequately define a
problem before rushing to judgment”. Building up shared knowledge in order to
gather relevant distributed knowledge of a problem is therefore a crucial issue.
Logical Information Systems (LIS) are based on Logical Concept Analysis
(LCA), an extension of Formal Concept Analysis (FCA). In a previous work [5],
Camelis, a single-user logical information system, has been shown useful to sup-
port serene and fair meetings. This paper shows how Abilis, a LIS web server that
implements OnLine Analytical Processing (OLAP [3]) features, can be applied
to help build shared knowledge among a group of skilled users.
The presented case study gathered a research team to decide on a publication
strategy. Starting from 1041 call for papers, each team member on his own
preselected a set of conferences matching his own focus of interest. The union of
individual preselections still contained 63 conferences. Then, participants met for
an hour and a half and collectively built a shared set of 42 target conferences. For
each conference, the team shared a deep understanding of why it was relevant.
The team could sketch a publication planning in a non-conflictual way.
Kolfschoten, de Vreede and Briggs have classified collaboration tasks into 16
collaboration patterns [12]. The contribution of this paper is to give evidences
that LIS can significantly support three of these patterns which are important as-
pects of decision making, namely Generate/Gathering, Clarify/Building Shared
Understanding and Reduce/Filtering. Firstly, the navigation and filtering capa-
bilities of LIS were helpful to detect inconsistencies and missing knowledge. The
updating capabilities of LIS enabled participants to add objects, features and
links between them on the fly. As a result the group had a more complete and
relevant set of information (Generate/Gathering pattern). Secondly, the com-
pact views provided by LIS and the OLAP features helped participants embrace
the whole required knowledge. The group could therefore build a shared under-
standing of the relevant information which was previously distributed amongst
the participants (Clarify/Building Shared Understanding pattern). Thirdly, the
navigation and filtering capabilities of LIS were relevant to quickly converge on
a reduced number of target conferences (Reduce/Filtering pattern).
In the following, Section 2 briefly introduces logical information systems. Sec-
tion 3 describes the case study. Section 4 gives detailed arguments to support the
claim that logical information systems help build up shared knowledge. Section 5
discusses related work.
2 Logical Information Systems
Logical Information Systems (LIS) [7] belong to a paradigm of information re-
trieval that combines querying and navigation. They are formally based on a
logical generalization of Formal Concept Analysis (FCA) [8], namely Logical
Concept Analysis (LCA) [6]. In LCA, logical formulas are used instead of sets
of attributes to describe objects. LCA and LIS are generic in that the logic
is not fixed, but is a parameter of those formalisms. Logical formulas are also
used to represent queries and navigation links. The concept lattice serves as the
navigation structure: every query leads to a concept, and every navigation link
leads from one concept to another. The query can be modified in three ways: by
formula edition, by navigation (selecting features in the index in order to modify
the query) or by examples. Annotations can be performed in the same interface.
Camelis3 has been developed since 2002; a web interface, Abilis 4 , has recently
been added. It incorporates display paradigms based on On-Line Analytical Pro-
cessing (OLAP). Instead of being presented as a list of objects, an extent can be
partitioned as an OLAP cube, namely a multi-dimensional array [1].
3
see http://www.irisa.fr/LIS/ferre/camelis/
4
http://ledenez.insa-rennes.fr/abilis/
3 The Case Study
The reported case study gathered 6 participants, including the 3 authors, 4
academics and 2 PhD students. All participants were familiar with LIS, 4 of them
had not previously used a LIS tool as a decision support system. The objective
was to identify the publishing strategy of the team: in which conferences to
submit and why. This has not been a conflictual decision, the group admitted
very early that the set of selected conferences could be rather large provided
that there was a good reason to keep each of them.
One person, the facilitator, spent an afternoon organizing the meeting and
preparing the raw data as well as a logical context according to the objective. She
collected data about conference call for papers of about a year, related to themes
corresponding to the potential area of the team, from WikiCFP, a semantic wiki
for Calls For Papers in science and technology fields 5 . There were 1041 events:
conferences, symposiums, workshops but also special issues of journals.
Then every participant, on its own, spent between half an hour to two hours
to browse the context, update it if necessary and preselect a number of con-
ferences (Section 3.1). The group met for one hour and a half. It collaborately
explored the data and selected a restricted set of conferences (Section 3.2). After
the meeting, every participant filled a questionnaire. The context used for the
case study can be freely accessed 6 .
3.1 Distributed Individual Preselection and Update
When the context was ready, every participant was asked to preselect a set of
conferences that could be possible submission targets. The instruction was to be
as liberal as wanted and in case of doubt to label the conference as a possible
target.
During this phase, each of the academics preselected 20 to 30 conferences and
each of the PhD students preselected around 10 conferences. Each participant
had his own “basket”. There were overlappings, altogether 63 conferences were
preselected. Participants also introduced new conferences and new features, for
example, the ranking of the Australian CORE association 7 (Ranking), and the
person expected to be a possible first author for the target conference (Main
author).
Figure 1 shows a state of Abilis during the preselection phase. LIS user in-
terfaces give a local view of the concept lattice, centered on a focus concept. The
local view is made of three parts: (1) the query (top left), (2) the extent (bottom
right), and (3) the index (bottom left). The query is a logical formula that typi-
cally combines attributes (e.g., Name), patterns (e.g., contains "conference"),
and Boolean connectors (and, or, not). The extent is the set of objects that are
matched by the query, according to logical subsumption. The extent identifies
5
http://www.wikicfp.com/cfp/
6
http://ledenez.insa-rennes.fr/abilis/, connect as guest, load Call for papers.
7
http://core.edu.au/index.php/categories/conference rankings
Fig. 1. Snapshot of Abilis during preselection: a powerful query
the focus concept. Finally, the index is a set of features, taken from a finite sub-
set of the logic, and is restricted to features associated to at least one object in
the extent. The index plays the role of a summary or inventory of the extent,
showing which kinds of objects there are, and how many of each kind there are
(e.g., in Figure 1, 8 objects in the extent have data mining as a theme). In the
index, features are organized as a taxonomy according to logical subsumption.
The query area (top left) shows the current selection criteria: (Name contains
"conference" or Name contains "symposium") and not (Name contains
"agent" or Name contains "challenge" or Name contains "workshop")
and (Theme is "Knowledge Discovery" or Theme is "Knowledge
Engineering" or Theme is "Knowledge Management"). Note that the query
had been obtained solely by clicking on features of the index (bottom left). Let
us describe how it had been produced. Initially there were 1041 objects. Firstly,
opening the Name ? feature, the participant had noticed that names could con-
tain “conference” or “symposium” but also other keywords such as “special
issue”. He decided to concentrate on conferences and symposiums by clicking
on the two features and then on the zoom button. The resulting query was
(Name contains "conference" or Name contains "symposium") and there
were 495 objects in the extent. However, the displayed features under Name ?
showed that there were still objects whose name in addition to “conference” or
“symposium” also contained “agent”, “challenge” or “workshop”. He decided to
filter them out by clicking on the three features then on the Not button then on
the zoom button. The resulting query was (Name contains "conference" or
Name contains "symposium") and not (Name contains "agent" or Name
contains "challenge" or Name contains "workshop") and there were 475
objects in the extent. He opened the Theme ? feature, clicked on the three sub-
features containing “Knowledge”, then on the zoom button. The resulting query
is the one displayed on Figure 1 and there are 48 objects in the displayed extent.
In the extent area (bottom right), the 48 acronyms of the selected conferences
are displayed. In the index area, one can see which of the features are filled for
these objects. The displayed features have at least 1 object attached to them.
The number of objects actually attached to them is shown in parentheses. For
example, only 14 of the preselected conferences have an abstract deadline. All
of them have an acronym, a date of beginning, a date of end, a date for the
paper deadline, a name, some other (not very relevant) information, as well as
at least a theme and a town. The features shared by all selected objects have
that number in parentheses (48 in this case). For the readers who have a color
printout, these features are in green. The other features are attached to only
some of the objects. For example, only 16 objects have a ranking attached to
them: 4 core A, 6 core B, 2 core C, 1 ‘too recent event’, 4 unknown (to the Core
ranking).
One way to pursue the navigation could be, for example, to click on Ranking ?
to select the conferences for which the information is filled. Alternatively, one
could concentrate on the ones for which the ranking is not filled, for example
to fill in this information on the fly for the conferences which are considered
interesting.
Another way to pursue the navigation could be, for example, to notice that
under the Theme ? feature, there are more than the selected themes. One can
see that among the selected conferences, one conference is also relevant to the
Decision Support Systems theme. One could zoom into it, this would add and
Theme is "Decision Support Systems" to the query ; the object area would
then display the relevant conference (namely GDN2011).
3.2 Collaborative Data Exploration, Update and Selection
The group eventually had a physical meeting where the current state of the
context was constantly displayed on a screen.
Using the navigation facilities of Abilis, the conferences were examined by
decreasing ranking. Initially, the group put in the selection all the A and A+
preselected conferences. After some discussions, it had, however, been decided
that Human Computer Interaction (HCI) was too far away from the core of
the team’s research. Subsequently, the HCI conferences already selected were
removed from the selection. For the conferences of rank B, the team decided
that most of them were pretty good and deserved to be kept in the selection.
For the others, the group investigated first the conferences without ranking and
very recent, trying to identify the ones with high selection rate or other good
reasons to put them in the selection. Some of the arguments have been added
into the context. Some others were taken into account on the fly to select some
conferences but they seemed so obvious at the time of the discussion that they
were not added in the context.
Figure 2 shows the selection made by the group at a given point. In the
extent area, on the right hand side, the selected objects are partitioned according
to the deadline month and the anticipated main author thanks to the OLAP
like facilities of Abilis [1]. Instead of being presented as a list of objects, an
extent can be partitioned as an OLAP cube, namely a multi-dimensional array.
Assuming object features are valued attributes, each attribute can play the role
of a dimension, whose values play the role of indices along this dimension. Users
can freely interleave changes to the query and changes to the extent view.
The query is SelectionLIS02Dec and scope international. Note that
the partition display is consistent with the query. When the group added and
scope international to the query, the national conferences disappeared from
the array.
Some conferences, absent from WikiCFP have been entered on the fly at that
stage (for example, ICCS 2011 - Concept). Not all the features had been entered
for all of them. In particular, one can see in the feature area that only 28 out
of 29 had been preselected. Nevertheless, the group judged that the deadline
month, the potential main author and the ranking were crucial for the decision
process and added them systematically. It is easy to find which objects do not
have a feature using Not and zoom, and then to attach features to them.
Fig. 2. Snapshot of Abilis during collaborative data exploration: a partition deadline
month/mainAuthor
One can see that there are enough opportunities for each participant to pub-
lish round the year. One can also see at a glance where compromises and decisions
will have to be made. For example, PC will probably not be in a position to pub-
lish at IDA, ISSTA, KDD and ICCS the same year. Thanks to this global view
PC can discuss with potential co-authors what the best strategy could be.
A follow up to the meeting was that participants made a personal publication
planning, knowing that their target conferences were approved by the group.
4 Discussion
In this section, we discuss how the reported case study provides evidences that
LIS help keep the group focused (Section 4.1) and that LIS also help build up
shared knowledge (Section 4.2). As already mentioned, participants filled up a
questionnaire after the meeting. In the following, for each item, we introduce
the arguments, we present a summary of relevant parts of participant feedbacks,
followed by an analysis of the features of LIS that are crucial for the arguments.
4.1 Logical Information Systems Help Keep the Group Focused
It is recognized that an expert facilitator can significantly increase the efficiency
of a meeting (see for example [2]). A study made by den Hengst and Adkins [10]
investigated which facilitation functions were found the most challenging by
facilitators around the world. It provides evidences that facilitators find that
“the most difficult facilitation function in meeting procedures is keeping the group
outcome focused.”
In our case study, all participants reported that they could very easily stay
focused on the point currently discussed thanks to the query and the consistency
between the three views.
As the objective was to construct a selection explicitly identified in Abilis
by a feature, the objective of the meeting was always present to everybody
and straightforward to bring back in case of digression. Furthermore, even if the
context contained over a thousand conferences, thanks to the navigation facilities
of LIS, only relevant information was displayed at a given time. Therefore, there
was no “noise” and no dispersion of attention, the displayed information was
always closely connected to the focus of the discussion.
4.2 Logical Information Systems Help Build Up Shared Knowledge
Kolfschoten, de Vreede and Briggs have identified 6 collaboration patterns: Gen-
erate, Reduce, Clarify, Organize, Evaluate, and Consensus Building [12]. We
discuss in the following three of their 16 sub-patterns for which all participants
agreed that they are supported by Abilis in its current stage. For the other sub-
patterns, the situation did not demand much with respect to them. For example,
the decision to make was not conflictual, the set of selected conferences could
be rather large, there was, therefore, not much to experiment about “consensus
building.” The descriptions of the patterns in italic are from Kolfschoten, de
Vreede and Briggs.
Generate/Gathering: move from having fewer to having more complete and rel-
evant information shared by the group.
Before and during the meeting, information has been added to the shared
knowledge repository of the group, namely the logical context. A new theme,
important for the team and missing from WikiCFP, has been added: Decision
Support Systems. New conferences have been added into the context either by
individual participants in the preselection phase or by the group during the
selection phase. New features were added. For example, it soon appeared that
some sort of conference rankings was necessary. The group added by hand, for the
conferences that were selected, the ranking of the Australian Core association.
Some conferences were added subsequently, sometimes the ranking was not added
at once.
All participants acknowledged that the tool helped the group to set up a set
of features which was relevant and reflecting the group’s point of view.
The crucial characteristics of LIS for this aspect are those which enable in-
tegrated navigation and update.
Firstly, the possibility to update the context while navigating in it enables
participants to enhance it on the fly adding small pieces of relevant information
at a time. Secondly, for each feature, Abilis displays the number of objects which
have it. It is therefore immediate to detect when a feature is not systematically
filled. The query Not selects the objects that do not have the feature.
Users can then decide if they want to update them. Thanks to the query, as soon
as an object is updated, it disappears from the extent. Users can immediately
see what remains to be updated. Thirdly, updating the context does not divert
from the initial objective. Indeed, the Back button allows users to go back to
previous queries. Fourthly, the three views (query, features, objects) are always
consistent and provide a “global” understanding of the relevant objects. Lastly,
in the shared web server, participants can see what information the others had
entered. Hence each participant can inspire the others.
For the last aspect, the facilitator inputs were decisive. Participants reported
that they did not invent much, they imitated and adapted from what the facili-
tator had initiated. This is consistent with the literature on group decision and
negotiation which emphasizes the key role of facilitators [2].
Clarify/Building Shared Understanding: Move from having less to more shared
understanding of the concepts shared by the group and the words and phrases
used to express them.
Participants, even senior ones, discovered new conferences. Some were sur-
prised by the ranking of conferences that they had previously overlooked. Par-
ticipants had a much clearer idea of who was interested in what.
All participants found that the tool helped them understand the points of
view of the others.
The crucial characteristics of LIS for this aspect are those which enable to
grasp a global understanding at a glance. Firstly, the query, as discussed earlier,
helps keep the group focused. Secondly, the consistency between the 3 views
helps participants to grasp the situation. Thirdly, irrelevant features are not in
the index, the features in the index thus reflect the current state of the group de-
cision. Fourthly, the partitions à la OLAP sort the information according to the
criteria under investigation. Lastly, the shared web server enables participants
to know before the meeting what the others have entered.
Reduce/Filtering: move from having many concepts to fewer concepts that meet
specific criteria according to the group members.
Both at preselection time and during the meeting, participants could quickly
strip down the set of conferences of interest according to the most relevant
criteria.
All participants said that the filtering criteria were relevant and reflecting
the group’s point of view. They also all thought that the group was satisfied
with the selected set of conferences.
The crucial characteristics of LIS for this aspect are those of the navigation
core of LIS. Firstly, the features of the index propose filtering criteria. They
are dynamically computed and they are relevant for the current selection of ob-
jects. Secondly, the query with its powerful logic capabilities enables participants
to express sophisticated selections. Thirdly, the navigation facilities enable par-
ticipants to build powerful queries, even without knowing anything about the
syntax. Lastly, users do not have to worry about the consistency of the set of
selected objects. The view consistency of Abilis guaranties that all conferences
fulfilling the expressed query are indeed present.
This aspect is especially important. As claimed by Davis et al. [4], conver-
gence in meetings is a slow and painful process for groups. Vogel and Coombes [16]
present an experiment that supports the hypothesis that groups selecting ideas
from a multicriteria task formulation will converge better than groups working on
a single criteria formulation, where convergence is defined as moving from many
ideas to a focus on a few ideas that are worthy of further attention. Convergence
is very close to the Reduce/Filtering collaboration pattern. They also underline
that people try to minimize the effects of information overload by employing con-
scious or even unconscious strategies of heuristics in order to reduce information
load, where information overload is defined as having too many things to do at
once.
With their powerful navigation facilities, LIS enable to address a large num-
ber of criteria and objects with a limited information overload. Indeed, one can
concentrate on local aspects. The global consistency is maintained automatically
by the concept lattice.
5 Related work
Abilis in its current stage does not pretend to match up to operational group
support systems (GSS) which have a much broader scope. LIS, however, could be
integrated in some of the modules of GSS. For example, M eetingworksT M [13],
one of the most established GSS, is a modular toolkit that can be configured to
support a wide variety of group tasks. Its “Organize” module proposes a tree
structure to help analyze and sort ideas. That structure looks much like the
index of LIS. It can be edited by hand and some limited selection is possible.
The navigation capabilities of LIS based on the concept lattice are, however,
more powerful.
Concept analysis has been applied to numerous social contexts, such as so-
cial networks [15], computer-mediated communication [9] and domestic violence
detection [14]. Most of those applications are intended to be applied a posteri-
ori, in order to get some understanding of the studied social phenomena. On
the contrary, we propose to use Logical Concept Analysis in the course and as a
support of the social phenomena itself. In our case, the purpose is to support a
collaborative decision process. Our approach is to other social applications, what
information retrieval is to data mining. Whereas data mining automatically com-
putes a global and static view on a posteriori data, information retrieval (i.e.
navigation in and update of the concept lattice) presents the user with a local
and dynamic view on live data, and only guides users in their choice.
A specificity of LIS is the use of logics. This has consequences both on the
queries that can be expressed, and on the feature taxonomy. The use of logics al-
lows to express inequalities on numerical attributes, disjunctions and negations in
queries. In pure FCA, only conjunctions of Boolean attributes can be expressed.
Previous sections have shown how disjunction and negation are important to
express selection criteria. In the taxonomy, criteria are organized according to
the logical subsumption relation between them in pure FCA, criteria would be
presented as a long flat list. Logics help to make the taxonomy more concise and
readable by grouping and hierarchizing together similar criteria. The taxonomy
can be dynamically updated by end-users.
6 Conclusion
In this paper we have shown that a Logical Information System web server could
be used to support a group decision process consisting of 1) data preparation
2) distributed individual preselection and update and 3) collaborative data ex-
ploration, update and selection. We have presented evidences that the navigation
and filtering capabilities of LIS were relevant to quickly reduce the number of
target conferences. Secondly, the same capabilities were also helpful to detect
inconsistencies and missing knowledge. The updating capabilities of LIS enabled
participants to add objects, features and links between them on the fly. As a
result the group had a more complete and relevant set of information. Thirdly,
the group had built a shared understanding of the relevant information.
Acknowledgments The authors thank Pierre Allard and Benjamin Sigonneau for
the development and maintenance of Abilis. They thank Pierre Allard, Annie
Foret and Alice Hermann for attending the experiment and giving many insight-
ful feedbacks.
References
1. Allard, P., Ferré, S., Ridoux, O.: Discovering functional dependencies and associa-
tion rules by navigating in a lattice of OLAP views. In: Kryszkiewicz, M., Obiedkov,
S. (eds.) Concept Lattices and Their Applications. pp. 199–210. CEUR-WS (2010)
2. Briggs, R.O., Kolfschoten, G.L., de Vreede, G.J., Albrecht, C.C., Lukosch, S.G.:
Facilitator in a box: Computer assisted collaboration engineering and process sup-
port systems for rapid development of collaborative applications for high-value
tasks. In: HICSS. pp. 1–10. IEEE Computer Society (2010)
3. Codd, E., Codd, S., Salley, C.: Providing OLAP (On-line Analytical Processing)
to User-Analysts: An IT Mandate. Codd & Date, Inc, San Jose (1993)
4. Davis, A., de Vreede, G.J., Briggs, R.: Designing thinklets for convergence. In:
AMCIS 2007 Proceedings (2007), http://aisel.aisnet.org/amcis2007/358
5. Ducassé, M., Ferré, S.: Fair(er) and (almost) serene committee meetings with logi-
cal and formal concept analysis. In: Eklund, P., Haemmerlé, O. (eds.) Proceedings
of the International Conference on Conceptual Structures. Springer-Verlag (July
2008), lecture Notes in Artificial Intelligence 5113
6. Ferré, S., Ridoux, O.: A logical generalization of formal concept analysis. In:
Mineau, G., Ganter, B. (eds.) International Conference on Conceptual Structures.
pp. 371–384. No. 1867 in Lecture Notes in Computer Science, Springer (Aug 2000)
7. Ferré, S., Ridoux, O.: An introduction to logical information systems. Information
Processing & Management 40(3), 383–419 (2004)
8. Ganter, B., Wille, R.: Formal Concept Analysis: Mathematical Foundations.
Springer, Heidelberg (1999)
9. Hara, N.: Analysis of computer-mediated communication: Using formal concept
analysis as a visualizing methodology. Journal of Educational Computing Research
26(1), 25–49 (2002)
10. den Hengst, M., Adkins, M.: Which collaboration patterns are most challenging:
A global survey of facilitators. In: HICSS. p. 17. IEEE Computer Society (2007)
11. Kilgour, D.M., Eden, C.: Handbook of Group Decision and Negotiation, Advances
in Group Decision and Negotiation, vol. 4. Springer Netherlands (2010)
12. Kolfschoten, G.L., de Vreede, G.J., Briggs, R.O.: Collaboration engineering. In:
Kilgour and Eden [11], chap. 20, pp. 339–357
13. Lewis, L.F.: Group support systems: Overview and guided tour. In: Kilgour and
Eden [11], chap. 14, pp. 249–268
14. Poelmans, J., Elzinga, P., Viaene, S., Dedene, G.: A case of using formal con-
cept analysis in combination with emergent self organizing maps for detecting
domestic violence. In: Perner, P. (ed.) Advances in Data Mining. Applications and
Theoretical Aspects, Lecture Notes in Computer Science, vol. 5633, pp. 247–260.
Springer Berlin / Heidelberg (2009), http://dx.doi.org/10.1007/978-3-642-03067-
3 20, 10.1007/978-3-642-03067-3 20
15. Roth, C., Bourgine, P.: Lattice-based dynamic and overlapping taxonomies:
The case of epistemic communities. Scientometrics 69, 429–447 (2006),
http://dx.doi.org/10.1007/s11192-006-0161-6, 10.1007/s11192-006-0161-6
16. Vogel, D., Coombes, J.: The effect of structure on convergence activities using
group support systems. In: Kilgour and Eden [11], chap. 17, pp. 301–311