The SemSearchXplorer - Exploring Semantic Search Results with Semantic Visualizations.
Open Research Online
The Open University’s repository of research publications
and other research outputs
The SemSearchXplorer - exploring semantic search re-
sults with semantic visualizations
How to cite:
Ullmann, Thomas Daniel; Uren, Victoria and Nikolov, Andriy (2009).
ing semantic search results with semantic visualizations. In: AST 2009 Applications of Semantic Technologies,
located at the Informatik 2009, 2 October 2009, Lbeck, Germany.
The SemSearchXplorer - explor-
For guidance on citations see FAQs.
c ? Gesellschaft fr Informatik, Bonn
Version: Version of Record
Link(s) to article on publisher’s website:
Copyright and Moral Rights for the articles on this site are retained by the individual authors and/or other copy-
right owners. For more information on Open Research Online’s data policy on reuse of materials please consult
the policies page.
The SemSearchXplorer – Exploring Semantic Search
Results with Semantic Visualizations
Thomas Daniel Ullmann, Victoria Uren*, Andriy Nikolov
Knowledge Media Institute (KMi)
The Open University
Milton Keynes, United Kingdom
Abstract: SemSearchXplorer is a toolkit for the exploration of semantic data.
The goal is to lower user barriers to access information in semantic data
repositories. Therefore SemSearchXplorer supports the user in three respects:
(1) it supports querying of the semantic data with a keyword based approach, so
the users do not need to learn a semantic query language, (2) it helps users find
relevant results both by using semantic enriched information about the results
and semantic filter options to narrow down the set of results, and (3) it provides
information exploration capabilities
recommended by the system. Filtering of semantic search results helps to
narrow down the result set to a more manageable amount of information.
Besides searching for relevant information, facilities for the exploration of the
results help users to gain insight in the context of results. With several semantic
visualizations, we try to help users making sense of the raw data. Based on the
assumption that there is no single visualization that fits all exploration needs,
SemSearchXplorer recommends visualizations based on the selected
information of users.
through semantic visualizations
Keywords: semantic search, semantic visualization, information exploration.
* Present address: Department of Computer Science, Regent Court, 211 Portobello, University of Sheffield,
Sheffield, S1 4DP United Kingdom.
1 From SemSearch to SemSearchXplorer
Semantic search engines return semantic enriched results of backing data structure,
which appropriate visualized support the searching and exploration tasks of
knowledge workers. We can subsume two strands of research associated with
semantic search. Searching for ontologies (e.g. Watson [Aq08] or Swoogle [Di04])
and searching within semantic data. The focus of this work lies on the latter one. With
SemSearchXplorer, we extended the semantic search engine SemSearch [LUM06] to
a toolkit for exploring semantic data following the notion of the importance of
iterative and exploratory search modes to the usability of search systems [Ur07].
With the keyword-based query interface of SemSearch, users do not need to learn
semantic query languages, like SPARQL [PS08] or SeRQL1, to query semantic data.
SemSearch translates automatically users’ keywords into semantic queries. The
metaphor for searching semantic knowledge-spaces with SemSearch is similar to
keyword-based search engines, and thus familiar for users. SemSearch returns, as
query results, instances and related triples (subject-predicate-object). In contrast to
query results not based on semantic web technology, these triples contain structured
information reflecting the relational information of the repository. Every result
contains a view to the backing data. The advantage of semantic data over unstructured
data is that this result can serve as a starting point for the exploration of the context of
the information due to the interconnected nature of ontologies.
SemSearchXplorer makes use of this structured information. Its user interface is
based on the knowledge lenses metaphor [Ur08]. This means that it provides the users
with several lenses or views to the data repository. The type of lens used is dependent
on information selected by users to explore the context of the SemSearch query result.
A common approach to visualize ontologies is an indented list [Ka07]; this is a tree
view like visualization of the ontology, for example used in the Neon Toolkit [Ha08]
or Protégé [No01]. Indented lists are simple to implement and users are familiar with
this concept, because of its common use in web-search engines. The downside of list
visualizations is that they can only represent tree and not graph structures. Graph
structures would be more suitable to visualize the interconnected structure of semantic
search results. However, it is not only list visualizations that have certain strengths
and weaknesses. There is no single visualization that fits every user information
exploration need [Ka07]. With SemSearchXplorer, we provide a range of
visualization types for the backing data with the goal of providing appropriate lenses
or views based on content.
We made the design decision for a content-based recommendation because of the
requirement that the search engine must be able to work out-of-the-box. A solution
based on personal or collaborative recommendations would have the benefit of
considering user characteristics, but would have the drawback that such systems only
works after the system has gathered some information about the user (cold start
While SemSearch supports users with a keyword-based query interface,
SemSearchXplorer focuses on the exploration of the retrieved results. We support this
exploration process with several semantic visualizations. Semantic visualization, or
ontology-based visualization [FSH03], is part of information visualization [CMS99,
Wa00, Sp07] and focuses on visualizing classes, instances, properties, and their
multiple relations. It is a part of the top layer of the semantic web stack [Be00] - the
user interface and application layer. The aim of semantic visualization is to visualize
the underlying structure of semantic data. One vision of semantic visualization would
be to use the content characteristics of the semantic web specification for the
automatic creation of visualizations.
2 Overview of SemSearchXplorer
One of the primary goals of SemSearchXplorer is to support the users with their
knowledge working tasks. Therefore, we defined two general requirements:
Hide technical details of semantic technology and semantic visualizations
techniques from users
Enhance the information search and exploration experience for users.
To meet the first requirement SemSearchXplorer uses SemSearch, which allows using
a google-like keyword search. Users have no need to learn semantic query languages
to query for information of the semantic web. For the details of SemSearch see
For the second requirement, SemSearchXplorer adds on top of the layer architecture
of SemSearch a visualization layer (see fig. 1). This layer is built as a pipeline. The
query results of SemSearch are the starting point of the pipeline. Based on user
interaction with SemSearchXplorer the filter and recommendation units set the course
for either the amount of processed information (filtering) or the type of the presented
visualization (recommendation). According to the recommended visualization and the
filter options, the engine transforms query results into the data model for the
visualization (e.g. table, graph or tree data structures). Then we enrich the data model
with visualisation information (e.g. colour, size or layout information) to the visual
form. The last unit of the pipeline is the view. A graphical renderer plots the actual
view on the screen.
Based on the actions of the users, either a new query is sent to SemSearch, or if a
visualization does not need new information from the search engine, the SemSearch
query results are transformed in another data model on which the new view is based.
Figure 1: Layer architecture of SemSearchXplorer
SemSearchXplorer uses this architecture to provide knowledge workers with
automatically recommended visualizations to support them with their information
SemSearchXplorer is built as a Java Swing application. It is meant to be used as a
browser application using the Java Web Start Technology.
3 Supporting Information Exploration with SemSearchXplorer
The Semantic Web is a web of connected resources. Resources are connected through
predicates with other resources forming a graph of resources. This concept allows
users to start their information exploration process with a resource and then explore
the context, the related resources of the chosen starting resource. It is important for
the user to find the right starting point for his search. Once found, the user can explore
the context of the information quite easily due to the underlying semantic web
In many cases, the user does not now much about the information in the knowledge
base. For this case, it is convenient for the knowledge worker to type in keywords into
a search interface to view the results the query engine returns. SemSearchXplorer uses
the semantic search engine SemSearch, which provides a keyword based query
interface. Users start their information exploration process with keyword queries.
SemSearch transforms these keywords into semantic queries and returns a ranked
result set of the query. Based on this result set SemSearchXplorer supports the user
with different lenses or visualizations of the result set helping him to refine and
explore the information.
The query interface of SemSearchXplorer (see fig. 2) allows either to search for a
keyword or to specify the expected type (the subject) of search results. The later is
supported through the syntax “subject:keyword”. With Boolean operators like and/or
we can compose complex queries (for details see [LUM06]). For example a query
about news of projects has the following form: “news:project”.
Figure 2: Initial query interface of SemSearchXplorer
3.2 Refining the Result Set
SemSearchXplorer visualizes the query results as an ordered list (see fig. 3). A list
representation has the advantage that users can read very quickly through the text and
scan for information. The information needs little space to plot. The order of the set of
results is sorted according to the ranking mechanism of SemSearch. Each hit contains
the instance using the fragment part of an URI, or if available, the label information
(rdfs:label). In addition to the instance we provide context information consisting of
the class names and the relations to other instances, which belong to the query. This
makes it easier for users to search for relevant hits. Each instance also provides a
tooltip with statistical information about the result.
Figure 3: SemSearchXplorer query result page
Query result page with
For users is it important to be able to narrow down the result set to a set of interesting
results. There are tradeoffs between filters with a small number of general class items,
which allow a quick reduction of the result set, and filters with a large number of filter
items, which allow a fine-grained narrowing, but have the drawback that the user has
to select from a large set of filter possibilities. To allow filtering of the result set, we
implement three types of filters. The first filter uses the direct super-classes of an
instance but not all the other classes higher in the class hierarchy. Compared to the
second filter, the all class filter, which uses the whole class information of the class
hierarchy of the result set, the direct class filter generates less filter items, while the
all class filter produces more filter items allowing precise filtering. The third filter
uses the key concepts (KCE – Key Concept Extraction [PMA08]) of an ontology.
With KCE it is possible to extract the ‘best descriptors’ of an ontology. Therefore, the
algorithm tries to find the most important concepts of an ontology as a human expert
would do. By now, the KCE algorithm considers three concepts. Natural categories
(that are concepts that are information-rich in psycholinguistic sense), density
(concepts which are information-rich in an ontological sense), and popularity
(frequency of terms returned from a query to the Yahoo search engine). Although this
algorithm's goal is to generate automatic ontology summaries, we consider KCE as a
method applicable for filtering the result set according to the most meaningful items
of an ontology while ignoring classes that are less important. Although the algorithm
tries to maximize the coverage of the ontology, some concepts are not contained in
the filter. We add hits of classes not belonging to the key concepts or subclasses to a
new filter entry called “other”.
With these three types of filters, the user has effective mechanism to filter quickly to
relevant information. Once found, the user can start the exploration of the hit.
After the user selected a result, a new view of SemSearchXplorer shows the
exploration page (see fig. 4). By now, we support users with three types of
visualizations to explore the context of the selected hit: a graph, cluster and chart
Figure 4: SemSearchXplorer exploration page with graph visualization
Graph visualizations are suitable for representing the context of a result. In contrast to
a list-based visualization, a graph visualization is able to represent the connections
between classes and instances. The list representation does not reflect the relational
structure of the semantic result. Within our graph visualization instances are
connected with edges representing the properties of the semantic result. While graph
visualization helps users to understand the relations between resources, large graph
visualizations become cluttered or need a large space on the screen.
Our cluster visualization (see fig. 5) is used for showing larger amounts of
information. Information that is not necessary for the first exploration is hidden in a
cluster. If users need this information, they can expand the cluster. With this
technique, we can reduce the problem of cluttered visualizations to a certain degree.
As the graph visualization the cluster visualization tries to support users making sense
of the relational structure of the backing data.
Result exploration: Cluster-,
chart-, and graph-
Figure 5: SemSearchXplorer exploration page with cluster visualization
Features of the cluster visualization are:
The instance to explore is highlighted and central in the graph representation.
The object instances of the triple are ordered in a cluster with the name of the
property. A cluster can contain a “more” cluster. We order higher ranked
instances into the first level group and lower ranked instances into the
On the top left position of the cluster visualization we provide an overview
and below the overview a tree representation of the whole visualization.
However, not only could the information about the context be interesting for users but
also statistical information about the instance. This allows users to compare instance
according their relevance. For an example of our chart visualization (see fig. 6). The
outer right blue chart of the chart visualization has the meaning that 12 instances have
as line manger Enrico Motta.
Figure 6: SemSearchXplorer exploration page with chart visualization
For a detailed discussion of the cluster and chart visualization and the underlying
algorithms, see [FUN09].
The most notable feature of the exploration page is the recommendation component
consisting of a variable number of recommended visualizations. Until now, we have
focused on content-based recommendations of visualizations. The goal would be to
provide users the most suitable view for a particular set of data (consisting of
instances, classes and/or relations). This is quite an ambitious vision, but for the first
implementation of SemSearchXplorer we consider heuristics which serve as starting
point for future recommendations. The heuristics follow these simple rules:
If the number of properties and related instances is above a threshold, then
offer a cluster visualization.
If this number is below a threshold, then offer a graph visualization.
If filtering influences this number significantly, then provide cluster or graph
visualization according to the threshold.
For chart visualizations, we need three clusters of data (x and y-axis, and the
values to plot). Therefore, if we can generate three clusters of data, then offer
a chart visualization.
Based on the chosen entities of the user SemSearchXplorer recommends the user
visualizations, which are able to visualize the selected information. As seen in figure
4 the recommendation of visualization is represented in iconic form below the
SemSearchXplorer logo. From left to right the icons are ordered according to which
of the possible visualizations the system predicts are more suitable.
4 Further prospects
The goal of SemSearchXplorer is to help people find relevant information more
quickly and to help them to explore the context of this information. For the case that
users do not exactly know what they are looking for, semantic search (SemSearch)
supports the user in finding a suitable starting point for their exploration and with the
visualization layer of SemSearchXplorer semantic visualizations help to explore the
context of the result.
SemSearchXplorer recommends visualizations according to the information the user
provides and the capability of the visualizations to make use of the provided
information. Thus, the decision is more a technical one, than based on user
experience. Further research will examine what questions the users want to answer
with SemSearchXplorer and what visualizations will help them doing that.
Another aspect concerns questions of user trust in information. The question is, do
people trust the visualized results more, if they can reproduce how the results are
generated and what sources of data are involved in the process. An explanation
component could help to unveil the mechanisms of the result generation. This
component provides more information than the actual info box of SemSearchXplorer,
which only shows basic statistical information of the query results.
After the user entered the query, a list-based representation of the visualization is
generated. The advantage of lists is that users are accustomed to read quickly through
linear text. This representation works well, if the user is only interested in the highest
ranked query results. However, if the user wants to get an overview of the whole
result set another visualization, which provides an overview of all hits at once, would
be more suitable.
With SemSearchXplorer and its software architecture we have an extensible testbed to
examine these outlined user centred and technical questions.
We wish to thank Fawad Nazir for his work on the cluster and graph visualization,
and Miriam Fernandez for her valuable comments on the search interface. In addition,
we want to thank for the helpful comments of the anonymous reviewers. This work
was funded by the X-Media project sponsored by the European Commission as part
of the Information Society Technology (IST) programme under EC Grant IST-FP6-
[Aq08] d'Aquin, M.; Sabou, M.; Motta, E.; Angeletou, S.; Gridinoc, L.; Lopez, V.; Zablith,
F.: What can be done with the Semantic Web? An Overview of Watson-based
Applications, 5th Workshop on Semantic Web Applications and Perspectives, SWAP
2008, Rome, Italy, 2008.
[Be00] Berners-Lee, T.: Semantic
[CMS99] Card, S. K.; Mackinlay, J. D.; Shneiderman, B.: Readings in information
visualization: using vision to think. Morgan Kaufmann Publishers, San Francisco,
[Di04] Ding, L.; Finin, T.; Joshi, A.; Pan, R.; Scott Cost, R.; Peng, Y.; Reddivari, P.; Doshi,
V. C.; Sachs, J.: Swoogle: A Search and Metadata Engine for the Semantic Web.
Proceedings of the 13th ACM Conference on Information and Knowledge
Management, November, 2004.
[FSH03] Fluit, C.; Sabou, M.; Harmelen, F.: Ontology-based Information Visualization. In
(Geroimenko, V.; Chen, C.; eds.): Visualizing the Semantic Web, pp 36-48.London:
[FUN09] Nazir, F.; Uren, V.; Nikolov, A.: Algorithms for Generating Ontology Based
Visualizations from Semantic Search Results, accepted paper for the Workshop
MoViX ‘09, 2009.
[Ha08] Haase, P.; Lewen, H.; Studer, R.; Tran, T.; Erdmann, M.; d’Aquin, M.; Motta, E.: The
NeOn Ontology Engineering Toolkit. WWW 2008, April 21-25, 2008, Beijing,
[Ka07] Katifori, A; Halatsis, C; Lepouras, G.; Vassilakis, C.; Giannopoulou, E.: Ontology
Visualization Methods – A survey. ACM Computing Survey, Vol. 39, No 4, October
2007; pp. 10.1-10.42.
[LUM06] Lei, Y.; Uren, V.; Motta, E.: SemSearch: A Search Engine for the Semantic Web. , In
(Staab, S.; Svatek, V.; eds.): Managing Knowledge in a World of Networks, 15th
International Conference, EKAW 2006, Podebrady, Czech Republic, October 2-6,
2006, Proceedings. Lecture Notes in Computer Science 4248 Springer 2006; pp. 238-
[No01] Noy, N. F.; Sintek, M.; Decker, S.; Crubezy, M.; Fergerson, R. W.; Musen, M. A.:
Creating Semantic Web Contents with Protégé-2000. IEEE Inteligent Systems,
March/April 2001, pp. 60-71.
[PMA08] Peroni, S., Motta, E., d'Aquin, M.: Identifying key concepts in an ontology through
the integration of cognitive principles with statistical and topological measures, Third
Asian Semantic Web Conference, Bangkok, Thailand, 2008
Web – XML2000. W3 Talks,
[PS08] Prud'hommeaux, E.; Seaborne, A.: SPARQL Query Language for RDF. W3C
Recommendation. January, 2008. http://www.w3.org/TR/rdf-sparql-query/, [2009-04-
[Sp07] Spence, R.: Information Visualization: Design for Interaction. Prentice Hall, 2007.
[Ur07] Uren, V.;Lei, Y.; Lopez, V.; Liu, H.; Motta, E.; Giordanino, M.: The usability of
semantic search tools: a review. The Knowledge Engineering Review, Vol. 22:4,
2007; pp 361-377.
[Ur08] Uren, V.; Franz, T.; Cimino, P.; Lei, Y.; Dadzie, A.; Gilardoni, L.: X-Media
Deliverable 4.4. Requirements for Knowledge Sharing Tools in Phase 2. Technical
Report of the X-Media project, 2008.
[Wa00] Ware, C.: Information Visualization: Perception for Design. Morgan Kaufmann
Publishers Inc, 2000.