=Paper=
{{Paper
|id=Vol-1304/STIDS2014_T04
|storemode=property
|title=Comprehension of RDF Data Using Situation Theory and Concept Maps
|pdfUrl=https://ceur-ws.org/Vol-1304/STIDS2014_T04_MoskalEtAl.pdf
|volume=Vol-1304
|dblpUrl=https://dblp.org/rec/conf/stids/MoskalKU14
}}
==Comprehension of RDF Data Using Situation Theory and Concept Maps ==
https://ceur-ws.org/Vol-1304/STIDS2014_T04_MoskalEtAl.pdf
Comprehension of RDF Data Using Situation
Theory and Concept Maps
Jakub J. Moskal Mieczyslaw M. Kokar Brian E. Ulicny
VIStology, Inc. Northeastern University VIStology, Inc.
Framingham, MA 01701, USA Boston, MA 02115, USA Framingham, MA 01701, USA
Email: jmoskal@vistology.com Email: m.kokar@neu.edu Email: bulicny@vistology.com
Abstract—The amount of RDF data available on the Web has Number of published Linked Data Datasets
been increasingly growing over the past few years. Developing
300
and fine-tuning SPARQL queries in order to sift through the data
may be a very challenging task for human operators who need to
quickly make sense of large graphs. In addition, often multiple
queries need to be issued in order to gather and understand the 225
context (relevant facts) for the explanation of the query. Thus,
the challenge is not only to answer the query, but also to provide
context, so that the analyst can easily comprehend what the data 150
is actually conveying.
This paper describes results of an investigation of the possibil-
ity to apply key aspects of Situation Theory, and its ontological 75
realization in the Situation Theory Ontology, to simplify and
abstract large RDF data sets, given a focus query from the
analyst. In this approach, the query results are presented as 0
concept maps. The approach was successfully implemented as a
20
20
20
20
20
20
20
20
20
20
20
20
prototype, although this paper does not include a description of
07
07
07
07
08
08
08
09
09
09
10
11
-0
-1
-1
-1
-0
-0
-0
-0
-0
-0
-0
-0
5-
0-
1-
1-
2-
3-
9-
3-
3-
7-
9-
9-
the tool.
01
08
07
10
28
31
18
05
27
14
22
19
Survey Date
I. I NTRODUCTION
Development of intelligence products in various domains, Fig. 1. Number of datasets that have been published in Linked Data format
e.g., business or military, requires sifting through tremen- between 2007 and 2011 [4].
dously large amounts of data, most of which so far is in
an unstructured (or semi-structured) form (text reports, web
pages). This constitutes a very high challenge to the analyst information is still very difficult to analyze. To illustrate the
who performs this kind of task. While the analyst has in problem, consider an example of analyst query about a gang-
mind an idea of the focus of the inquiry, the focus may exist related activity:
only in the analyst’s head and thus cannot be supported by What were the circumstances of Richard H. Barter’s death?
a computer-based tool. One way for the analyst to tell the
computer what is being looked for is to issue a search query, Such a query can be expressed in SPARQL query language
e.g., using keywords. However, the tools that support keyword- [5] using the DESCRIBE query and the FILTER command that
based text search will return documents (or pointers to) that makes use of regex pattern matching to extract all the facts
contain the words; the analyst still needs to do the hard work that are related to “Richard H. Barter”. Even though DBpedia
of reviewing the plethora of documents returned. Another way had only one resource (“Richard H. Barter”) that is directly
is to first use a text processing tool that will analyze the related to the query, the query returns more than 25 other
documents, extract entities and relations identified in those resources that are one way or another related to this resource.
documents and represent them in a structured language, e.g., DESCRIBE queries return RDF graphs and in order to analyze
Resource Description Language (RDF) [1], and then analyze such an answer the analyst would have to go over all of the
the resulting formal representation using an appropriate query links and nodes and decide which of them are relevant.
language. An example of the development in this domain is Now the question is how to present the result of the query
the idea of Linked Data [2], which has resulted, among others, to the analyst? One of the formats for visual representation
in a quite large knowledge base called DBpedia [3]. of complex information structures that has been proved quite
In fact, DBpedia is just one of the numerous open datasets successful in various uses, including knowledge structuring,
that have been published in RDF format. As the chart in learning and even knowledge creation, is the representation
Figure 1 shows, the number of such datasets has been rapidly called Concept Map [6], [7], [8]. However, as discussed later
growing in the recent years. Unfortunately, the RDF structured in the paper, concept maps that are direct representations of
25
Concept Maps
represent
Organized Knowledge
Concepts is comprised of Propositions
connected used to
using form
Linking Words
Fig. 2. Example of a concept map, representing the notion of concept maps
itself [12].
Fig. 3. Small RDF graph, returned by a SPARQL DESCRIBE query,
visualized as a concept map using the IHMC CMap tool.
RDF graphs can also become quite complex and thus difficult
to comprehend.
The problem addressed in this paper is the transformation The concept map of Figure 2 shows five concepts and four
of RDF to concept maps so that the resulting concept map is propositions, where one of the concepts (Concept Maps) is a
relevant to a specific analyst query, includes the appropriate “meta-concept”, since it represents the notion of a concept map
context, and is presented in a more abstract form than the itself. This map is only a fraction of a larger map, which shows
original RDF so that it is easy to comprehend. Our approach the key features of concept maps [12]. Note that different look
is to use key aspects of Situation Theory of Barwise and and feel styles can be applied to both concepts and linking
Perry [9], as extended and formalized by Devlin [10], map words, e.g., different colors for different types of concepts.
our problem to this theory and implement algorithms for Now returning to our example of the above described query,
constructing concept maps based on such a framework. In this the CMap tool can load the answer to the SPARQL query and
work, we used the Situation Theory Ontology (STO) [11] that convert it to a concept map (c.f. Figure 3), however, the analy-
we developed earlier. sis of the map is still not that easy. One of the main reasons for
The rest of this paper is organized as follows. In Section II this difficulty is the fact that concept maps generated in this
we briefly overview concept maps. In Section III we briefly way will contain too many concepts and relationships, many
discuss why Situation Theory is a good candidate for the of them not relevant to the query. (Note: Clearly, Figure 3 is
solution. Then in Section IV we show how we can represent not readable. The sole purpose of this figure is to show the
analyst queries in the STO ontology. This is followed by the complexity of such concept maps.) One way to simplify the
discussion of domain inference in Section V and situation presentation would be to display just a small portion of the
reasoning in Section VI. Section VII describes the derivation concepts and relationships. However, this operation needs to
of (possibly) multiple contexts related to a query. Section VIII be performed very carefully so that important facts, without
then discusses how the contexts are simplified in order to which the analyst would not be able to understand the answer
make the derived concept maps easier to comprehend. Finally, to the query, are not omitted. Furthermore, the answer might
Section IX presents the conclusions of the paper and suggests include too detailed information, which clutters the global
some of the possible directions for future research. conceptual picture and defeats the purpose of the concept map.
Hence, a fine balance between the simplicity and the amount
II. C ONCEPT M APS of information must be kept in order to allow the analyst to
quickly explore and understand the data.
A concept is defined [8] as a perceived regularity or pattern
designated by a label. Propositions are statements about some
III. S ITUATION T HEORY
object or event in the universe, either naturally occurring
or constructed. Propositions contain two or more concepts Situation Theory is “a set of mathematically-based tools to
connected using linking words or phrases to form a meaningful analyze, in particular, the way context facilitates and influences
statement. Sometimes these are called semantic units, or units the rise and flow of information” [10]. Situation theory came
of meaning. about from the attempts to formalize Situation Semantics
Concept maps (c.f. Figure 2) include concepts (represented – reasoning about common sense and real world situations
as boxes) and relationships between concepts (propositions) [9]. As postulated by Barwise and Perry, situations are first-
indicated by connecting lines linking pairs of concepts. Words class objects, i.e., they have their own existence, can stand in
in the boxes represent concept names, while words on/above relation with other objects (including other situations) and can
the lines represent relationships between two concepts. Since have their own attributes.
concepts and properties are the building blocks of RDF, RDF In situation theory, information about a situation is ex-
graphs can be seen as concept maps. The CMap tools from pressed in terms of infons written as:
IHMC can be used to provide graphical representations of
RDF graphs as concept maps [8]. ⌧ R, a1 , . . . , an , 0/1
26
where R is an n-place relation and a1 , . . . , an are objects supportedInfon* Situation
appropriate for R. Since situation theory is multi-sorted, the focalIndividual*
word “appropriate” means that the objects are of the types ElementaryInfon relevantRelation*
appropriate for a given relation. The last item in an infon is focalRelation*
relevantIndividual*
relation Individual
the polarity of the infon. Its value is either 1 (if the objects
stand in the relation R) or 0 (if the objects don’t stand in the
Relation hasAttribute*
relation R). polarity
hasAttribute*
To capture the semantics of situations, situation theory impliedByRule*
provides a relation between situations and infons. This re- Attribute
lationship is called the supports relationship which relates
Rule hasAttributeValue*
a situation with the infons that “are made factual” by the hasDimensionality
situation. Given an infon and situation s the proposition
“s supports ” is written as: isa
Polarity Value Dimensionality
s |= .
Fig. 4. Top-level classes in the STO ontology.
The relation between a situation (in the world) and a
representation of the situation (in a formal framework) is
relative to a specific agent. It is the agent who establishes
such a link. This link is defined by connections that link
entities in the world to formal constructs of the situation- types of properties: relation (to point to the relation, R that
theoretic framework. These connections are not part of the the infon represents), and anchor (not shown in Figure 4) to
formal theory. One refers to situations within a formal theory point to the arguments of R. Polarity in STO is represented
by using abstract situations, although the qualifier “abstract” explicitly, i.e., positive assertions correspond to polarity 1 and
is often dropped in most discussions of situation theory. An negative assertions correspond to polarity 0.
abstract situation is then a collection of infons supported by a One of the possible alternatives to Situation Theory that we
specific situation. looked at was the FrameNet approach [14]. FrameNet is based
In our approach we mapped key aspects of Situation Theory on a theory of meaning called “frame semantics” derived from
to Situation Theory Ontology (STO). The top-level classes the work of Fillmore et al. (cf. [15]). The basic idea is that the
of STO are shown in Figure 4. The details of this ontology meanings of most words can best be understood on the basis
were described elsewhere [13]. Here we just mention that of a semantic frame: a description of a type of event, relation,
the main idea behind this ontology is to capture the concept or entity and the participants in it.
of “situation” (the Situation class serves this purpose). An
individual s0 of Situation that corresponds to a situation s in While this idea seems to be close to that of the Situation
the real world, serves as the root to the description of the Theory (ST) semantics, the latter has a number of advantages
situation s. The abstract situation associated with s0 is the that make ST a better match for this particular problem. (1)
context; it holds all the facts that are relevant to the situation, ST grounds meaning in the world rather than in the language.
s. Other classes included in Figure 4 include Relation (to This allows for the development of situation types that have
represent relations that individuals - instances of the class meaning in the physical reality (e.g., battlefield, ships, missiles
Individual - are involved in), Attribute (to represent attributes and so on), and not just in the syntax of the human language.
of both individuals and situations), Value and Dimensionality (2) Unlike in more pure logic-based semantics, meaning in
of the attributes, Rule (to represent rules for inferring higher- ST is provided by partial views of the world, not all possible
arity relations) and Polarity (to represent the values of Polarity; worlds. This gives an advantage of being able to specify
the only instances of this class are 1 and 0). views of the world (situations) that are globally inconsistent,
It is important to stress here that STO approximates Situa- but locally consistent. This will allow analysts to specify
tion Theory by capturing the supports relation with a entails situation types that, when taken together, are inconsistent.
(or derives) relation, `, between the collection of infons This capability allows the deference of the resolution of
represening a situation and the infon representing a query [13]. inconsistencies to the interactions with the world, rather than
Moreover, information is not represented in the form of infons. trying to develop a consistent set of types (an impossible state
Instead, STO uses OWL and/or rules to represent knowledge to achieve) before anything is utilized. (3) Situations in ST
about situations, i.e., abstract situations are captured by OWL are first-class objects, i.e., they not only stand in relations
sentences. However, as shown in Figure 4, STO includes the with other situations, but can have their own attributes and
class ElementaryInfon. The sole role that ElementaryInfon properties. (4) In ST the meaning of a declarative sentence is
plays in STO is to capture the focus of specific situations. a relation between utterances and described situations, which
I.e., queries (expressed in natural language) are formalized is exactly what is needed for a solution to our problem —
as instances of this class. ElementaryInfon resembles the developing concept maps that support the understanding of
structure of the infon in Situation Theory and thus has two answers to specific analyst’s questions (queries).
27
IV. R EPRESENTING Q UERIES of clauses, each being a triple (predicate, subject, object). The
In our approach, the essence of the textual version of analyst syntax of BVR is conceptually compatible with RDF. This
queries needs to be extracted and mapped to to the ontology. kind of rules are easy to write and interpret; the only problem
Since situations are explicitly represented in STO, the mapping is that it is verbose. For this reason, BVR offers an abbreviated
of the queries to STO has to be consistent with the intent of this syntax [16].
ontology. In particular, since the intent is to connect a query The activities involved in the answering of analyst queries
with a context (which in STO is captured by a situation), as and creating concept maps that constitute the answers, is
well as to ensure that the relevant facts are included in the shown in Figure 5. The following sections describe each of
context, queries were mapped to the class of ElementaryInfon these activities in more detail.
and to a specific situation type.1 For instance, the answer to V. D OMAIN I NFERENCE
the query whose textual representation is The first step in the processing of an analyst query is to
“Did an insurgent visit a weapons cache”?” run the inference on the supplied RDF data and infer implicit
facts about the domain (Step 1 in Figure 5). Since RDF does
can be captured by InsurgentWeaponsCacheSituation (a sub-
not provide strong axioms for inference, the RDF data can
class of the Situation class), defined in OWL as follows:
be augmented with additional axioms expressed in OWL and
InsurgentWeaponsCacheSituation ⌘ Situation and (supporte- rules. OWL was the preferred choice, but if for some axioms
dInfon some (ElementaryInfon and (anchor1 some Insurgent) it was not expressive enough, axioms were added in the form
and (anchor2 some WeaponsCache) and (relation value visit))) of BVR rules.
For instance, for the SynCOIN dataset [17] used in our ex-
Answering such a query would involve inferring whether periments, examples of domain-specific axioms are definitions
the current knowledge base supports the conclusion that of object properties associate and madeTransactionsWith, both
there is a situation individual that is a member of the class of which were defined as sub-properties of the transitive and
InsurgentWeaponsCacheSituation. Note that the above def- symmetric isConnectedTo property (left side of Figure 6). An
inition assumes that the domain-specific ontology used in example of the use of these axioms is shown on the right
this query extends STO with some classes (e.g., Insurgent, side of the figure. Assuming that only John has been to a
WeaponsCache) and relations (e.g., visit). weapons cache, and that Mary is the only known insurgent,
Unfortunately, OWL is not sufficient enough to express if the analyst issues a query “Which known insurgents are
some types of queries. For instance, the following query cannot connected to people who have been to a weapons cache?”,
be expressed in OWL alone: the system should produce a map that includes Mary and
John. In addition, the map should also include Bob and the
“Which insurgents spied on a relative?”
relationships between all individuals, in order to fully represent
The reason for this is that one needs to refer to variables, which the context. Without Bob in the result, it is not obvious how
are not supported by OWL. In particular, the intent of this Mary and John are actually connected.
query is to identify only those insurgents who spied on their While the process of adding domain-specific axioms needs
own relatives, not just any insurgents who spied someone’s to be done manually, it is part of the knowledge engineering
relatives. In such cases one needs to use rules. For instance, task, which is expected to be performed for each domain
using the STO, the query above could be expressed as the of application. Obviously, automatic ways of generating such
following rule: axioms are desirable, but this was not part of this investigation.
Situation(s) ^ ElementaryInfon(i) ^ Object(a1) ^ Object(a2) ^ In our case, we arbitrarily decided which axioms to include.
Relation(spiedOn) ^ supportedInfon(s,i) ^ anchor1(i, a1) ^ an-
chor2(i, a2) ^ relation(i, spiedOn) ^ Insurgent(a1) ^ Person(a2) Ontology isConnectedTo
Example
^ relative(anchor1, anchor2) ! RelativeSpySituation(s) isConnectedTo
Transitive
Symmetric associate
Mary Bob isConnectedTo
Such rules can be captured in SPARQL 1.1 (using INSERT madeTransactionWith
to assert new facts) or in an inference engine-specific language associate madeTransactionWith isConnectedTo
like BaseVISor’s RDF-based BVR [16]. For the ease of use, John
since it was already the language in which some of the domain
axioms were expressed (discussed below), BVR was chosen Fig. 6. Property taxonomy and example of its use. The blue lines represent
as the query language. In BVR, rules are defined within a rule implicit, inferred facts.
base with each rule consisting of a body element and a head
element (which can occur in either order). The name attribute VI. S ITUATION R EASONING
can be used to assign a name to a rule base or rule. The heads
Once the domain inference is complete and all implicit
and the bodies use the triple syntax, i.e., each rule consists
domain facts are asserted in the knowledge base, individuals
1 In OWL a query about some individuals can be viewed as a class, i.e., a of a situation type that corresponds to the query, as well as
collection of those individuals that satisfy the definition of the class. relations among them, can be found (Step 2 in Figure 5).
28
RDF augmented with Situations matching the query
Original RDF Data
Inferred Facts detected in the RDF
DOMAIN SITUATION
1 INFERENCE 2 REASONING
Detected Situations augmented
Concept Maps
Situations with context
CONTENT
3 DERIVATION 4 SIMPLIFICATION
Fig. 5. The process of transforming RDF data into comprehensible concept maps described in this paper. The situations detected in the RDF graph correspond
to answers to an analyst query, which gives a focus for the produced concept maps.
To begin with, situation type definitions need to be an- Now the reasoner can infer the situation types of the situation
alyzed — both those that are defined in pure OWL and individuals.
those that are defined in rules (see Section IV for details).
The main focus here is to extract the relations used in the VII. D ERIVATION OF C ONTEXTS
definition of the situation types. For instance, for the Insur- At this point, the answers consist of the anchors and the
gentWeaponsCacheSituation type, the system should extract relations used in the situation definitions. For instance, for
the visit relation. Similarly, for RelativeSpySituation, it would the weapons cache query and the axioms shown in Figure 6,
extract the spiedOn relation. Getting this information from given that John has been to a weapons cache, the system
OWL definitions is trivial, since we know the structure of would return a basic concept map including Mary, John and
the definition of situation types, which use the notion of isConnectedTo, but not include Bob and his relationships with
ElementaryInfon, which in turn explicitly uses the object them, which would explain why Mary is actually connected to
property relation. It gets more complicated with the situation John. Hence, the next processing step is to derive the context
types defined in rules. In our experiments, the BVR rule for the answer, i.e., find all individuals and relations that are
files were processed with regular expressions in order to find relevant to the situation that represents the answer to the query.
the relations. In the future, the rules themselves could be This corresponds to step 3 in Figure 5. Recall that “context”
formally represented in OWL and the solution could avoid the means an abstract situation, as described earlier in the paper.
use of regular expressions. Once the relations from situation The main idea is that context is the description of a situation,
type definitions are known, the process of asserting situation including all the relevant individuals and the relevant relations
individuals is as follows: among the individuals. All of this (the context) is captured by
• For each relation rel that is part of a situation type: the relevant facts, i.e., facts that assert which individuals and
– For each pair of individuals a1 and a2 that are relations are relevant and what are the relations among the
associated with each other by the property rel: relevant individuals.
1) Assert that there is an individual s of RDF type For deriving context, we implemented a set of domain-
sto:Situation independent rules, that backtrack some of the OWL inference
2) Assert that there is an individual i of RDF type rules. For instance, if a relation that is relevant to the query is
sto:ElementaryInfon, supported by situation s defined as a property chain, the individuals and relations that
3) Assert the following facts: (i anchor1 a1 ), (i form the chain are inferred to be relevant as well. Similarly,
anchor2 a2 ) and (i relation rel) if a relevant relation is defined as a super-property of another
29
property that holds between two relevant individuals, it is also Similarly to the previous steps, for this purpose we de-
inferred to be relevant. At the time of writing, the set of the veloped a number of domain-independent rules that remove
context derivation rules is not complete, i.e., not every OWL redundant facts. As an example, the following algorithm
inference rule that produces new facts has a corresponding describes the rule that removes from a situation’s context those
relevance derivation rule. Also, some rules might produce facts properties whose sub-properties, holding between the same
that are not necessary to explain a situation to the analyst, thus individuals, are relevant, yet not necessary:
producing some “noise”. Such issues are on our agenda for • For a situation s, and a query q, if s satisfies the query:
future work. – For every relation r1 and r2 both relevant to s, if r1
As an example, the following describes one of the relevance is a sub-property of r2 :
derivation rules related to the transitive properties in OWL2 :
⇤ For every two facts (i1 r1 i2 ) and (i1 r2 i2 ) that
• For a situation s, and a query q, if s satisfies the query: are both relevant to s:
– For every fact (i1 rel i2 ) relevant to s and an 1) Remove (i1 r2 i2 ) from the context of s.
individual i3 , if rel is a transitive property and if
Back to the weapons cache example, based on the above rule
(i1 rel i3 ) and (i3 rel i2 ) are facts asserted in the
applied to the graph in Figure 7, the system would remove the
knowledge base:
two isConnected links between Bob and the other two people,
1) Add (i1 rel i3 ) and (i3 rel i2 ) as facts relevant since they both provide redundant information. The associate
to s. and madeTransactionWith properties are more specific and
Figure 7 shows how derivation rules can be applied in clearly explain the context for the original query.
the weapons cache example, given the axioms in Figure 6. The resulting concept map could use different graphical
First, based on the above rule applied to isConnectedTo, styles when rendering concepts and links, in order to distin-
the inference engine would infer that the individual Bob is guish the query answer itself from its context. This approach
also relevant and should be part of the context (Figure 7b). gives the analyst a quick focus on the most important concepts
Moreover, using a different derivation rule, the reasoner would in the graph, but also provides the context without cluttering
infer that associate and madeTransactionWith are also relevant, the answer.
because they are sub-properties of a relevant property and hold
between relevant individuals (Figure 7c). IX. C ONCLUSION
Note that not only individuals and properties are asserted The main objective of the research described in this paper
as relevant to a situation, but entire facts (triples) are also was to investigate the possibility of using the ideas from
asserted as such. It is not sufficient to just list the individuals Situation Theory (Barwise, Perry and Devlin), and its onto-
and properties without showing the associations between them. logical realization in the Situation Theory Ontology, to the
In our experiments, we used the notion of OWL annotation task of simplifying and abstracting concept maps, provided
properties in order to annotate facts as relevant to specific as RDF graphs, so that they are easier to comprehend by an
situation individuals. Since OWL does not support reasoning analyst while still preserving the semantics of the original
over annotation properties, the only way to implement such representation. This paper covers only some of the aspects
reasoning is to use rules. As we mentioned earlier in the paper, of this investigation. In particular, it shows (by example)
our preference was to use OWL reasoning first and add rules how an analyst’s query can be mapped to an ontological
only out of necessity. representation, what it takes to derive facts that are relevant
to the query, and how to represent such facts in graphical
VIII. S IMPLIFICATION OF C ONCEPT M APS form (both with and without auxiliary facts that provide an
One can easily see that as a result of context derivation explanation to the analyst of how they were derived). This
reasoning, the number of relevant facts for each situation might investigation ended with a prototype tool (not included in this
grow fast and if converted into a concept map, it could look paper) for generating, displaying and manipulating concept
quite convoluted (compare Figure 7a with Figure 7c). More maps in order to improve their comprehensibility. The next
importantly, it would most likely include redundant facts. For logical task for this research is to evaluate the tool on a
instance, Figure 7c shows that Mary and Bob are associated representative number of queries and datasets and assess the
using two properties isConnectedTo and associate, although approach with respect to its completeness and the strength
the former is just a generalization of the latter. of the rules used for the simplification of the query results.
In order to make such resulting concept map less cluttered, In particular, such an evaluation would require human-in-the-
and thus easier to comprehend, we need to remove facts loop, i.e., the involvement of the analysts performing analyses
that are relevant to a situation, but that are not necessary to of situational awareness in their domains.
comprehend the graph. We call this step context simplification ACKNOWLEDGMENT
and it corresponds to step 4 in Figure 5.
This work was performed under Office of Naval Research
2 Note that it is not important whether the facts on which the rule operates contract N00014-14-P-1081 “Concept Maps from RDF (Re-
were derived or asserted by the user. source Description Framework)”. Any opinions, findings and
30
Initial Fact Transitive Property Derivation Sub-Property Derivation Sub-Property Simplification
isConnectedTo isConnectedTo
Mary Mary associate Bob
Mary Bob Mary associate Bob
isConnectedTo isConnectedTo madeTransactionWith
isConnectedTo madeTransactionWith isConnectedTo
isConnectedTo
isConnectedTo
John
John John John
(a) (b) (c) (d)
Fig. 7. Example of the context derivation and simplification of a query answer rendered as a concept map.
conclusions or recommendations expressed in this material [17] J. L. Graham, D. L. Hall, and J. Rimland, “A coin-inspired synthetic
are those of the authors and do not necessarily reflect the dataset for qualitative evaluation of hard and soft fusion systems,” in
Information Fusion (FUSION), 2011 Proceedings of the 14th Interna-
views of the Office of Naval Research. The authors would also tional Conference on. IEEE, 2011, pp. 1–8.
like to thank the anonymous reviewers who provided many
constructive suggestions for improving the presentation and
for future research directions.
R EFERENCES
[1] W3C, “RDF semantics. W3C recommendation 10 february, 2004,”
Feburary 2004. [Online]. Available: http://www.w3.org/TR/2004/
REC-rdf-mt-20040210/
[2] T. Heath and C. Bizer, Linked Data: Evolving the Web into a Global
Data Space. Synthesis Lectures on the Semantic Web: Theory and
Technology. Morgan & Claypool Publishers, 2011.
[3] DBPedia-Community, “DBpedia,” September 2014. [Online]. Available:
http://dbpedia.org/About/
[4] R. Cyganiak and A. Jentzsch, “The linked open data cloud diagram,”
http://lod-cloud.net/, September 19 2011.
[5] J. Pérez, M. Arenas, and C. Gutierrez, “Semantics and complexity of
sparql,” in The Semantic Web-ISWC 2006. Springer, 2006, pp. 30–43.
[6] J. D. Novak, “Concept maps and vee diagrams: Two metacognitive tools
for science and mathematics education,” Instructional Science, vol. 19,
pp. 29–52, 1990.
[7] E. Plotnik, “Concept Mapping: A Graphical System for Understanding
the Relationship between Concepts,” ERIC Clearinghouse on Informa-
tion and Technology Syracuse NY, vol. ED407938, 1997.
[8] J. D. Novak and A. Cañas, “The Theory Underlying Concept Maps
and How to Construct and Use Them,” cmap.ihmc.us/publications/
researchpapers/theorycmaps/.
[9] J. Barwise and J. Perry, Situations and Attitudes. Cambridge, MA: MIT
Press, 1983.
[10] K. Devlin, “Situation theory and situation semantics,” in Handbook of
the History of Logic, D. M. Gabbay and J. Woods, Eds. Elsevier, 2006.
[11] “Situation Theory Ontology.” [Online]. Available: http://vistology.com/
onts/2008/STO/STO.owl
[12] J. D. Novak and A. J. Cañas, “The theory underlying concept maps
and how to construct and use them,” Institute for Human and Machine
Cognition, Tech. Rep. Technical Report IHMC CmapTools 2006-01 Rev
2008-01, 2008. [Online]. Available: http://cmap.ihmc.us/publications/
researchpapers/theorycmaps/theoryunderlyingconceptmaps.htm
[13] M. M. Kokar, C. J. Matheus, and K. Baclawski, “Ontology-based
situation awareness,” Information fusion, vol. 10, no. 1, pp. 83–98, 2009.
[14] C. F. Baker, C. J. Fillmore, and J. B. Lowe, “The Berkeley FrameNet
project,” in Proceedings of COLING/ACL, 1998.
[15] C. J. Fillmore, “Frame semantics,” in Linguistics in the Morning Calm.
Seoul, Korea: Hanshin Publishing Co., 1982, pp. 111–137.
[16] C. J. Matheus, K. Baclawski, and M. M. Kokar, “BaseVISor: A triples-
based inference engine outfitted to process RuleML and R-Entailment
rules,” in Rules and Rule Markup Languages for the Semantic Web,
Second International Conference on, 2006, pp. 67–74.
31