Mapping the Knowledge Artifact Terrain
A Quantitative Resource for Qualitative Research
Federico Cabitza and Angela Locoro
Dipartimento di Informatica, Sistemistica e Comunicazione,
Universit
`
a degli Studi di Milano Bicocca, Viale Sarca, 336, 20126, Milano, Italy
Keywords:
Knowledge Artifact, Situativity, Objectivity, Natural Language Processing, Polarity Detection.
Abstract:
In this paper, we present a method by which to build a metaphorical map of a portion of the scholarly literature
along conceptual dimensions that have been previously characterized in terms of positive, negative and neutral
terms. The method allows to “locate” scholarly works in this space, according to multiple criteria, like the
definitions that they contain; the relevant concepts that can be extracted by means of a content analysis; and
relevant passages that researchers can extract in studying their content. The resulting maps are not representa-
tional, nor trying to extract any objective essence of a scientific contribution. Rather, they are resources for the
qualitative research, review and interpretation of literature sources. As such, these maps are “knowledge arti-
facts” in themselves, as they visualize, so to say, the interpretation of a set of works by qualitative researchers,
and allow to build a visual comprehension of topological and qualitative relationships between the considered
literature contributions. We applied the method to the case of the “knowledge artifact” literature and report
the main results in this paper.
1 INTRODUCTION
A Knowledge Artifact (KA) is any artifact built to
support knowledge-related processes. This purposely
generic definition allows to cover a broad spectrum of
instances of this concept, which nevertheless present
many differences and mirror different perspectives to-
ward what knowledge is and how it can be supported.
In a recent qualitative literature survey (Cabitza
and Locoro, 2014), the authors accomplished an ex-
tensive review of the heterogeneous scholarly contri-
butions that had focused that far on the concept of
Knowledge Artifact. This review resulted in an inter-
pretative and bottom-up framework by which to char-
acterize single instances of a KA in terms of two op-
posite and complementary design dimensions: objec-
tivity and situativity. In that work objectivity was de-
noted as the dimension characterizing the KAs that
are more oriented to a model-driven and Artificial In-
telligence (AI) approach to knowledge management.
Conversely, situativity was considered the dimension
characterizing those KAs that adopt a more construc-
tivist, practice- and collaboration-oriented approach
to knowledge support. It is obvious that clear-cut dis-
tinctions are useful only for theoretical and analyti-
cal purposes, and that in reality both dimensions are
present at different degrees in each Knowledge IT Ar-
tifact (KITA). The above-mentioned literature review
was aimed at discovering this two-dimensional mix
by a qualitative analysis that was carried out by the
researchers manually.
An ideal continuation of this approach may be the
application of a more systematic method by which to
extract a sort of polarity of the literature sources un-
der consideration along some dimensions of interest,
like the two mentioned above; and then to represent
these sources in the metaphorical space defined by
these conceptual dimensions, which are assumed to
be orthogonal and independent.
Thus this paper can be seen as a follow-up of the
contribution mentioned above in that: 1) it proposes
a method by which to build a knowledge artifact sup-
porting the study of a body of scholarly contributions;
and 2) also, by applying this method in a case study
focusing on the KA literature, it sheds light on the
more recent and relevant contributions from this body
of literature.
The method by which the conceptual space men-
tioned above would be populated is based on the con-
tent that the researchers extract from the literature
sources during their analysis, and takes also in con-
sideration specific lexicons (i.e., lists of words) that
444
Cabitza, F. and Locoro, A..
Mapping the Knowledge Artifact Terrain - A Quantitative Resource for Qualitative Research.
In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 3: KMIS, pages 444-451
ISBN: 978-989-758-158-8
Copyright
c
2015 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
Table 1: The objectivity (above) and situativity (below) lexicons.
algorithm, analytical, artifact, autonomous, autonomy, bi, body, business, capture, categorize, codification, codify, com-
bination, communicate, communication, complete, computational, concept, conceptual, correct, crm, datawarehouse,
decidable, decision, determinist, deterministic, discrete, document, dss, encode, engineer, engineering, exchangeable,
expert, explicit, externalization, externalize, externalizing, factual, fix, formal, formalism, formality, functionalist, hard-
coded, hard-code, independent, independently, information, intelligence, is, ka, km, knowledge, management, map,
metadata, minimalistic, minimalist, model, nomothetic, objective, objectivist, objectivistic, objectivity, olap, ontol-
ogy, order, outcome, passive, positivist, positivistic, predictive, prescribe, prescriptive, problem, procedural, procedure,
process, processual, procedural, rational, record, regulation, repository, represent, representable, representation, repre-
sentational, retrieve, semantic, semi-formal, solve, specify, static, store, structure, symbolic, top-down, transfer, valid,
validity, validate.
action, actionable, activity, affordance, agency, articulate, articulation, artifact, augment, awareness, beyond, body,
bottom-up, brainstorming, cad, chaotic, co-worker, co-create, co-creation, collaborate, collaboration, collectivity, col-
lectively, communicate, communication, community, constructivism, constructivist, constructivistic, consumer, con-
text, contextual, continuous, convey, cooperate, cooperation, cop, creative, cultural, cybernetic, decision, document,
emerge, emergentist, enable, evolve, experience, externalize, embody, fit, flexibility, flexible, fluid, groupware, holistic,
human-embodied, human-embody, incomplete, informal, information, innovation, input, integrate, interact, interac-
tion, interative, intermediary, internalization, interpretation, interpretive, interpretable, ka, knowledge, learn, learner,
learning, local, locality, malleability, malleable, manipulate, mediate, mediation, negotiable, nominalism, nominal-
ist, nominalistic, organization, others, partial, perform, performance, personalization, practice, pragmatic, pragmatist,
pragmatically, presence, problem, producer, product, proxy, reconcile, reconcilement, relational, result, retrieve, share,
situate, situation, situational, situativity, skill, social, socialization, sociotechnical, socio-technical, solve, stakeholder,
subjective, subjectivity, support, synergistic, tacit, team, think, thinking, training, transfer, undecidable, underspec-
ify, understand, understanding, unpredictive, unstructured, unstructure, usable, use, user-driven, utilization, vehicle,
voluntarist, voluntaristic, word, working.
the researchers have previously prepared for each di-
mension at hand (in our case, objectivity and situativ-
ity). These lexicons contain so called polar terms be-
cause their content is compared to the extracted con-
tent in order to measure the degree of polarity of a
paper with respect to each dimension, i.e., its degree
of dimension-ness.
As anticipated above, this space should be con-
sidered a knowledge artifact in itself for the follow-
ing reasons. Points in this space would not merely
“represent” literature sources, but rather help the re-
searcher look at, in a way, her reified interpretation
of those contributions, and be supported in getting a
visual comprehension of the literature of her inter-
est (in our specific case of the body of literature re-
garding the KA concept). The space and the objects
therein located would then support reflective insight,
collaborative discussion, and discovery, also, e.g., of
proximities, affinities, alignments, trends that can be
found among the analyzed sources (obviously still on
a metaphorical level). The maps that result from the
application of our method can then be seen as re-
sources for the qualitative interpretation of a body of
literature, as this latter is processed in terms of lin-
guistic prevalence and polarity. We will see how this
is accomplished in the next Section; then in Section 3,
we will validate the method by applying it to the KAs
research, and we will give some examples of how the
considered literature can be mapped giving visual evi-
dence of the intrinsic diversity of the scholarly contri-
butions and their possible interpretations. At last, we
will be back to the map metaphor again in Section 4,
which will close the paper.
2 METHOD
The method that we aim to propose for the visualiza-
tion of scientific content along the dimensions of ob-
jectivity and situativity
1
considers three intertwined
aspects that contribute in making a literature source
valuable: i) the definition aspect, represented in terms
of all of the sentences that in a paper give an ex-
plicit definition of a KA; ii) the design-oriented as-
pect, represented in terms of the sentences that in the
paper describe the functionalities or the main require-
ments motivating the design of a KA in the paper; iii)
the theoretical aspect, represented in terms of a list
of nominal categories that researcher can extract by
trying to understand the underlying assumptions that
drove the authors of a paper in discussing the defini-
tions as well as the design aspects of a KA. In par-
ticular, terms for describing the theoretical aspect can
result from any technique of content analysis and in-
terpretative paradigm aimed at the construction of a
theory, ontology or model, through the analysis of the
paper’s content, and can be seen as framework meta-
keywords that can be added to the paper through an
1
The reader should mind that the method is intended to
be general with respect to the dimensions of analysis, and
that we applied it to these two dimensions for the sake of
example only.
Mapping the Knowledge Artifact Terrain - A Quantitative Resource for Qualitative Research
445
Figure 1: Visualization of the component vectors for, respectively, the definition, design, and theoretical aspects of the three
papers examined.
annotation process, to enrich it or support its interpre-
tation.
As anticipated in the Introduction, the method is
proposed as an interpretative tool, which could sup-
port expert scholars in getting a multimodal (i.e., both
textual and visual) comprehension of the conceptual
dimensions of the works they consider, flanking tradi-
tional techniques of content analysis that require both
reading and extracting meaningful material out of the
research papers under consideration.
Among the most popular applications of polar-
ity detection in written (and spoken) texts there are
“opinion mining” and “sentiment analysis” (Pang and
Lee, 2002). These approaches aim to “integrate
emotional aspects in natural language understand-
ing” (Cambria and Hussasin, 2015), in order to extract
information on the dimension of the “sentiment” of
people, that is their feeling or mood, being these vot-
ers, consumers, community-members or simply citi-
zens. For instance, the main approaches to sentiment
analysis are machine-learning and dictionary-based
ones (Rice and Zorn, 2013). The latter approach ex-
ploits a list of polar terms to give a measure of the
emotional valence of a text (either positive or nega-
tive), being it either a whole document, or a portion
of it, like just a single sentence.
In general, common techniques used to measure
a dimension of a text, like its “sentiment”, encom-
pass the representation of texts by means of word vec-
tors and word vector comparison by means of docu-
ment classifiers (Pang et al., 2002). However, “mining
opinions and sentiments from natural language [. . . ]
is an extremely difficult task as it involves a deep un-
derstanding of most of the explicit and implicit, regu-
lar and irregular, syntactical and semantic rules that
are proper of a language” (Cambria and Hussasin,
2015). Moreover, for long texts and generic vocab-
ularies “such approaches turns critically on the qual-
ity and comprehensiveness with which the dictionary
reflects the sentiment in the texts to which it is ap-
plied” (Rice and Zorn, 2013). For these reasons, a po-
larity detection approach would more probably result
suitable whenever applied to a specific domain, with
a specific domain vocabulary created by hand by do-
main experts, and applied to very short domain texts
or group of sentences. And as a means to further com-
prehension and interpretation of multiple texts, rather
than as an end in itself.
In this paper, we try to generalize the approach
of sentiment analysis to the analysis of any dimen-
sion that can be characterized by a vocabulary of
dimension-relevant (either positive or negative) terms
to extract and evaluate the polarity and valence (along
that dimension) of any text. In our specific case,
we are interested in potential expressions of objec-
tivity and situativity within a scientific KA-related
contribution, instead of positive or negative emotions
in generic content. We have applied the resulting
method to the case of the body literature regarding
both “representational KAs” and “socially-situated
KAs” (Cabitza and Locoro, 2014).
More precisely, in our method we represent strings
(i.e., sets of words, or 1-grams) in terms of vec-
tors along a dimension, that is elements in a mono-
dimensional vector space characterized by: a direc-
tion (i.e., a polarity, which we indicate as a coefficient
α); and a magnitude, i.e., the degree of dimension-
ness of the string; as said above, a common example
of dimension is “sentiment”, which can be either neg-
ative or positive, as well as situativity and objectivity,
as in the case study reported in Section 3.
We model the vector magnitude in terms of the
product of two components: β and γ. The former
is the valence of the predominat polarity within the
string with respect to the other polarity (within the
same string): therefore it is a sort of relative degree of
polarization of the string. The latter (γ) is the degree
of absolute polarity of the string, that is how much it
is polarized with respect to the total length. In other
words, we take the importance of the most relevant
lexicon within a text (γ), and then we weight it accord-
ing to its relative importance with respect to the other
lexicon within a specific string (β). Because the vec-
tor magnitude is normalized with respect to the string
KITA 2015 - 1st International Workshop on the design, development and use of Knowledge IT Artifacts in professional communities and
aggregations
446
length, string vectors are defined on a continuous in-
terval [-1, 1] symmetric around the origin (which rep-
resents perfect neutrality).
Obviously, strings can be represented along more
than one dimension (e.g. (Mikolov et al., 2013)). In
this case, they are vectors in a multidimensional vec-
tor space, that is sum vectors, resulting from the com-
position of each dimension vector defined above.
As anticipated in Section 1, to illustrate our
method we consider just two particular dimensions:
objectivity and situativity. For each of these two di-
mensions we defined one lexicon, D, divided in two
partitions, D
+
and D
: these are unordered sets of
both neutral and positive/negative words (1-grams,
excluding stop-words), respectively, with respect to
the dimension at hand. Furthermore, for simplic-
ity’s sake, in this study we assume that these two di-
mensions are mutual opposites, that is S
Situativity
+
S
Ob jectivity
, and vice versa.
As hinted at in Section 1, our method also consid-
ers to extract three strings from each literature contri-
bution: the set of the definitions therein given by the
authors and identified as such by the researcher; the
set of codes that the researcher assigns to the paper,
following any hermeneutics technique; the set of the
relevant passages that the researcher extracts from the
paper. Each one of these strings are vectors in the bi-
dimensional vector space, or objectivity - situativity
plane.
Here we emphasize the main aim of the method
that is represented in the three Formulas 1, 2 and 3
to calculate polarity, valence and the string vector
magnitude, respectively. While a more syntactic ap-
proach would have been aimed at representing a re-
search contribution within the vector space by apply-
ing the method to the whole paper content, we aim
it at supporting qualitative research. Therefore, we
apply the algorithm represented in the formulas men-
tioned above to content that is produced by the qual-
itative researcher during the study of each literature
source. We then map the outputs of her study on the
definition, design and theoretical level in terms of a
cumulative vector (each vector associated with one
single paper) on the conceptual plane.
α =
(
1 if
|
S
D
+
|
<
|
S
D
|
+1 if
|
S
D
+
|
|
S
D
|
(1)
|
β
D
|
=
(
|
S
D
grt
|
|
(S
D
lwr
\(S
D
grt
S
D
lwr
))
|
|
S
D
+
S
D
|
if
|
S
D
+
|
6=
|
S
D
|
ε otherwise
(2)
γ
D
i
= β
D
i
·
|
S
D
grt
|
|
S
|
(3)
In Formula 1, α is the polarity coefficient; it is +1
if D
grt
is D
+
(the positive lexicon defined for dimen-
sion D), that is when the cardinality of D
+
is greater
than the cardinality of D
(the negative lexicon de-
fined for dimension D ); α is -1, otherwise, that is if
D
grt
is D
or the cardinality of D
is greater than the
cardinality of D
+
.
In Formula 2, S
D
+
is the set of occurrences of
the words belonging to D
+
, that has been found in
the string S (also, the set of word occurrences con-
noted either positively or neutrally along dimension D
within S). S
D
is the set of occurrences of the words
belonging to D
that has been found in the string
S (also, the set of word occurrences connoted either
negatively or neutrally along dimension D within S).
S
D
grt
and S
D
lwr
indicate which set among the two sets
mentioned above has greater and lower cardinality, re-
spectively, with respect to the other set. The ε param-
eter is an arbitrarily small constant (e.g., .1) to allow
for the visualization of null polarity. In our method,
if we detect a prevalence of either positive or nega-
tive words within the string S, we connote the neu-
tral words, that are the words contained in D
+
D
,
as either positive or negative, respectively. This sim-
ple context-based and majority-driven polarization of
neutral words is reasonable when the difference be-
tween the cardinality of S
D
+
and S
D
is much greater
than the number of neutral words defined in both lex-
icons, as it is often the case.
In Formula 3, γ
D
i
is the magnitude of the string
vector, that is the coordinate associated with valence
β
D
i
taken along the axis representing the i-th dimen-
sion. In calculating the cardinality of S, duplicate
words are considered as many times they are repli-
cated.
3 THE CASE STUDY
The method described in Section 2 was applied to the
case of the “knowledge artifact” literature, in order to
validate it and report the outcome of its application
to a typical qualitative research task: literature review
and study.
To this aim, we created the objectivity and sit-
uativity lexicons by hand; as anticipated above,
these lexicons encompassed positively related terms,
negatively-related ones in two distinct partitions, and
neutral words in both of them, i.e. words that can-
not be considered as either clearly positive or nega-
tive, but that nevertheless are relevant for the domain
at hand (e.g., “knowledge”). We then applied our
method of polarity detection to a selection of research
papers identified in (Cabitza and Locoro, 2014).
Mapping the Knowledge Artifact Terrain - A Quantitative Resource for Qualitative Research
447
Figure 2: Mapping of the considered papers in regard to their definition aspect within the objectivity-situativity plane.
To create the lexicons we proceeded as follows.
We considered 22 papers that in (Cabitza and Locoro,
2014) were found to contain one or more explicit def-
inition statements of the KA, and 3 more research pa-
pers from the 10 papers that in that literature survey
were found to be concerned with KA design. These
latter works are (Massey and Montoya-Weiss, 2006;
Salazar-Torres et al., 2008; Giunchiglia and Chenu-
Abente, 2009). The authors extracted the terms from
these primary sources independently, on the basis of
their perceived relevance and capability to connote
the concepts of objectivity and situativity (either pos-
itively or negatively).
The overall list of terms contained all of the terms
extracted by both of us, while terms extracted by only
one of the authors were reviewed jointly to decide for
its inclusion on the list. Then we classified the identi-
fied terms as either positive or negative (and of course
neutral), also in this case independently: inter-rater
agreement was assessed in terms of Cohen’s kappa,
and this process iterated each time after a short dis-
cussion on the term that had been classified differ-
ently. After a few iterations, we got sufficient agree-
ment for both the objectivity and situativity lexicons,
which are reported in Table 1. So far objectivity terms
are 103, whereas the situativity lexicon contains 145
terms. We recall here that, as part of our method, the
lexicons can be continuously refined, for any further
research purpose, also according to the papers consid-
ered and the raters involved.
The 22 papers mentioned above were used also
to validate our procedure of polarity detection and
visual mapping. In particular, we extracted the KA
definitions as bag-of-words, and processed them as
1-grams by means of a Python script that executed
the lemmatization of word tokens
2
and the removal
of stopwords
3
. We then computed the valence men-
tioned in Section 2 according to Formulas 2 and 1,
and computed the objectivity as well as the situativity
degrees by exploiting Formula 3.
Figure 2 depicts the visual mapping of the 22 pa-
pers according to their definitions of KA in terms of
situativity and objectivity coordinates, which result
from the application of our method
4
.
An example of insight that this kind of visualiza-
tion can suggest is how the literature has changed over
time: from Figure 2, by looking at the paper position
and their date of publication (marked in the point la-
2
We used the Wordnet lemmatizer of the nltk package
available at http://www.nltk.org/index.html
3
The stopwords list is availabe
at http://algs4.cs.princeton.edu/35applications/stopwords.txt
4
In Figure 2 we chose to plot only the resulting points,
and not the entire vector, for readability’s sake.
KITA 2015 - 1st International Workshop on the design, development and use of Knowledge IT Artifacts in professional communities and
aggregations
448
bel) one can observe that, in recent times, the number
of situativity-oriented works is increasing over time
(showing perhaps a growing interest or sensitivity to-
wards that approach?). Another example is the ob-
servation of the formation of “clusters” of papers in
particular areas of the space (e.g. the I, the II and
the IV quadrants), and with particular distance pro-
portions and shapes: e.g. the uniformity of spacing
between the papers located in the II quadrant; the mu-
tual proximity of the papers of the I quadrant; the
more scattered set of papers of the IV quadrant. This
can give some hints on the linguistic uniformity in the
definitions of the “representational” (hence more ob-
jective) KAs and, on the contrary, of the linguistic
variety in the definitions provided for the “socially-
situated” KAs. Also, a further look can be given
to see whether close papers are also describing sim-
ilar KAs, or whether the same researcher (or group
of researchers) has maintained the same theoretical
“stances” over time, and so on.
In general, with this kind of qualitative visualiza-
tion it is possible to inquire into the linguistic prop-
erties of the papers that the method has grouped to-
gether (spatially) to see whether they should be also
grouped conceptually to some extent (e.g., by writing
style, by theoretical stance, by objectivity and situa-
tivity attitude, and so on). As the researchers’ aims
can vary a lot, as well as their “mental models”, dif-
ferent “mappings” can be open to various interpreta-
tion and insights, according to the research scope and
aims that tap in this particular knowledge artifact.
For the 3 papers that focused on three specific KA
applications, we proceeded as follows. We created
three word sets, obtained by extracting: their defi-
nitions (as in the 22 papers mentioned above); the
relevant passages describing the concrete application
of the KA at hand; the keywords that the authors
agreed to apply to each of them. For this latter task,
we followed a hybrid approach (Fereday and Muir-
Cochrane, 2008), trying to apply the lexicon terms
first, but then also any other code we agreed upon,
after a short cycle of iterative revisions.
Applying the script mentioned above to these
three word sets, we obtained the component vectors
related, respectively, to the definition-, the theory-
and design-related aspects of each paper (see Fig-
ure 1 for a representation of each dimension as a
vector for each paper). This representation allows
to compare each of the relevant dimensions of a pa-
per, and to compare them with the same dimensions
of other papers, discovering for example that the pa-
per (Salazar-Torres et al., 2008) has a uniform di-
rection of the three dimensions of definition, the-
ory and design, whereas the paper (Giunchiglia and
Chenu-Abente, 2009) presents quite an opposite view
in terms of definitiorial vs theoretical stances; fi-
nally, the paper (Massey and Montoya-Weiss, 2006)
lays “somewhere in the middle”, presenting with a
more emphasized socially-situated stance along the
theoretical dimension with respect to the definito-
rial dimension. The design-related aspects of both
the papers (Giunchiglia and Chenu-Abente, 2009)
and (Massey and Montoya-Weiss, 2006) seem to go
in the opposite directions along the theoretical dimen-
sion, although the design aspect seems to be less tech-
nical in the latter work than in the former one, which
covers the technical details of the knowledge artifact
at hand for approximately half of its length.
By composing together these three vectors for
each of the papers under consideration, we obtained
their overall vector representation on the objectivity-
situativity vector space, as depicted in Figure 3.
This figure shows that our content analysis was
such that paper (Salazar-Torres et al., 2008) looks less
situativity-oriented than paper (Massey and Montoya-
Weiss, 2006), and that paper (Giunchiglia and Chenu-
Abente, 2009) looks much more objectivity-oriented
than the first two papers
5
. Figure 2 can also suggest
considerations about the relative (conceptual) prox-
imity and mutual alignment between works: (Salazar-
Torres et al., 2008) is maybe closer to (Massey
and Montoya-Weiss, 2006) than to (Giunchiglia and
Chenu-Abente, 2009), but even more importantly,
they look more “aligned” to each other.
4 DISCUSSION AND
CONCLUSIONS
Terrain is not a territory. For the dictionary, a ter-
rain is any piece of ground with reference to its phys-
ical character. It is therefore the element of a terri-
tory that human beings have to cope with. We intend
this term metaphorically to intend the scholarly land-
scape encompassing the diverse literature contribu-
tions regarding any concept, and in our specific case
the concept of “knowledge artifact”. In this light, we
intend the term terrain used in the title to be seman-
tically closer to the French terroir, a term that tra-
ditionally “refers to the complex ecology in which a
given vineyard is located [and that] evokes the unique
5
We believe that trying to distinguish if the papers ac-
tually are or rather look according to the researcher inter-
pretative lens would be out of the scope of this work. The
maps provided by our method, like those depicted in Fig-
ures 3 and 2 are given as complementary visual resources to
content analysis and literature interpretation.
Mapping the Knowledge Artifact Terrain - A Quantitative Resource for Qualitative Research
449
Figure 3: Visual mapping of the composite vectors for the three papers examined.
qualities of the soil, weather patterns, situated wine-
making practices, sunshine, and irrigation that yields
a particular and recognizable character to the wines
that result” (McNely and Rivers, 2014). If we take
the metaphor seriously, this means that the body of
literature contributions on a particular topic “exists
outside human control” (ibid.) but it also bears fruit
(like knowledge, insights, new ideas) only in the ac-
tual practice of the researchers that explore and, so to
say, harvest it.
Following the metaphor again: in order to exploit,
take possession of, and also orient themselves in, ter-
rains, human beings construct and use maps. One im-
portant point many cartography enthusiasts know well
is that maps do not necessarily depict or represent
(figuratively) terrains; obviously maps relate to ter-
rains, but they are rather “resources for action” (Such-
man, 2007), that is means to understand, explore and
exploit terrains. A typical practice where maps are
used is wayfinding: in this case, maps are just like
marked pathways in the wood, or signposts affixed at
relevant crossroads. Another practice is also contem-
plation: maps can be consulted just for the aesthetic
pleasure to find them accurate, complete, up-to-date,
clean and elegant. This should not be considered a
lazy activity, as maps can also act as triggers for de-
tecting relationships between terrain elements, as well
as to reflect on them and discuss them.
In this paper we have presented a way to map a
metaphorical terrain of a portion of the scholarly lit-
erature found to be related to a specific topic. This
terrain unfolds conceptually along discursive dimen-
sions, that is dimensions that are characterized in
terms of positive, negative and neutral terms. The
mapping method we devised, although simple, allows
for a cursory “locating” of scholarly works in this
space according to multiple criteria, like the defini-
tions that they contain; the relevant concepts that can
be extracted by means of a content analysis; and rele-
vant passages that researchers can extract in studying
their content. Some of the insights that researchers
may pull out from the visual representations given in
all of the Figures of this paper have been illustrated
in Section 3 as an outcome of our analysis and clas-
sification of the papers. Once again, we stress the
fact that the task of placing single contributions in the
terrain of interest (that is a discursive terrain) should
then not be taken as a task of representation of each
source’s position (even supposing such a thing really
exists or can be pinpointed in any metaphorical space)
within this landscape. Indeed, our method is highly
dependent on qualitative content analysis in regard to
all of its inputs: both the dimensional lexicons, and
the strings (set of words) representing each literature
source.
The resulting maps that our quantitative method
allows to build are not aimed (nor built) to represent
a body of works, nor to extract any objective essence
of a scientific contribution, if such a thing exists. Far
from it, those maps are intended as resources for the
interpretation of selected papers by the qualitative re-
searcher, as an aid to literature reviews to allow for
qualitative comparisons and evaluations, and a trig-
ger for discussion and idea exchange between schol-
ars about what they study.
That is why we claim that these maps are “knowl-
edge artifacts” in themselves; indeed, they visual-
ize, so to say, the interpretation of a set of works by
qualitative researchers, and allow them to build a vi-
sual comprehension of topological and qualitative re-
lationships between the considered literature contri-
butions.
In particular, we applied the method to the case of
the “knowledge artifact” literature. As such the paper
contributes in the literature regarding the concept of
knowledge artifact in regard to two main aspects.
KITA 2015 - 1st International Workshop on the design, development and use of Knowledge IT Artifacts in professional communities and
aggregations
450
First, we have defined pairs of lexicons (positive
and negative ones) for the dimensions of situa-
tivity and objectivity (of knowledge artifacts), as
these had been defined in (Cabitza and Locoro,
2014). These lexicons have been both produced
and reconciled manually, after a comprehensive
review of a set of relevant papers and an itera-
tive process of inter-rater agreement that involved
the authors. These vocabularies are offered to the
community of interested researchers to be pro-
gressively maintained, and to enable further re-
search on these topics along the same research
strand advocated in (Cabitza and Locoro, 2014).
Then, we propose an algorithm for the mapping
and visualization of arbitrary sets of words (either
directly taken or derived from the original liter-
ature sources) into the objectivity-situativity bi-
dimensional vector space. In this paper, to val-
idate the method we produced the resulting out-
puts:
1. the set of the main definitions of knowl-
edge artifact explicitly given by the authors of
22 papers selected from the review presented
in (Cabitza and Locoro, 2014);
2. the sets of definitions and relevant categories
that we extracted from three relevant papers
selected from the literature review mentioned
above;
3. the sets of all of the relevant design-oriented
passages that we extracted from each of these
papers, having in mind the concrete applica-
tions mentioned in each contribution.
Each paper has then be mapped in terms
of a graphical representation within a vector
space, by considering its definition-, theory-
and application-oriented aspects (respectively, the
word sets of its definitions, categories and rel-
evant passages). The vector-like representation
should be also appraised for the related affordance
of a “tension” and for evoking a “tendency” rather
than a mere position, which eludes any too rigid
pinpointing of the characteristics of a research
contribution.
If the interpretation of a set of literature sources
and the discussion of the related topics can be fos-
tered by looking at the vector space that we propose
to build with our method as a map for qualitative lit-
erature reviews, the main goals of our study would
be reached. In this case, our method could comple-
ment the study of scholarly sources, and facilitate the
qualitative researcher in extracting insights from the
literature.
REFERENCES
Cabitza, F., Locoro, A. (2014). ”Made with Knowledge” -
Disentangling the IT Knowledge Artifact by a Qual-
itative Literature Review. In Procs of KMIS 2014.,
64–75.
Cambria, E., Hussein, A. (2015). Sentic Computing: A
Common-Sense-Based Framework for Concept-Level
Sentiment Analysis Springer.
Fereday, J., & Muir-Cochrane, E. (2008). Demonstrating
rigor using thematic analysis: A hybrid approach
of inductive and deductive coding and theme devel-
opment. International journal of qualitative methods,
5(1), 80-92.
Giunchiglia, F. & Chenu-Abente, R. (2009). Scientific
Knowledge Objects v.1. Techreport DISI-09-006,
University of Trento.
Massey, A.P. & Montoya-Weiss, M.M. (2006). Unravel-
ing the temporal fabric of knowledge conversion: A
model of media selection and use. Mis Q, 30(1), 99–
114.
McNely, B. J., & Rivers, N. A. (2014, September). All of
the things: Engaging complex assemblages in com-
munication design. In Procs of the ACM International
Conference on The Design of Communication, p. 7.
Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., &
Dean, J. (2013). Distributed representations of words
and phrases and their compositionality. In Advances
in neural information processing systems (pp. 3111-
3119).
Pang, B., Lee, L. and Vaithyanathan, S. (2002). Thumbs
up?: sentiment classification using machine learning
techniques In Procs of the ACL-02 conference on
Empirical methods in natural language processing,
10:79–86.
Pang, B., Lee, L. (2008). Opinion mining and sentiment
analysis Foundations and trends in information re-
trieva, 2(1):1–135.
Turney, P.D. and Pantel, P. (2014). Corpus-based dictionar-
ies for sentiment analysis of specialized vocabularies
In Procs of NDATAD 2013.
Salazar-Torres, G. et al. (2008). Design issues for knowl-
edge artifacts. Knowledge-Based Systems, 21(8),
856-867.
Suchman, L. (2007). Human-machine reconfigurations:
Plans and situated actions. Cambridge University
Press.
Turney, P.D. and Pantel, P. (2014). From frequency to mean-
ing: Vector space models of semantics. J Artif Intell
Res, 141–188.
Mapping the Knowledge Artifact Terrain - A Quantitative Resource for Qualitative Research
451