=Paper=
{{Paper
|id=None
|storemode=property
|title=SemSor: Combining Social and Semantic Web to Support the Analysis of Emergency Situations
|pdfUrl=https://ceur-ws.org/Vol-747/paper4.pdf
|volume=Vol-747
}}
==SemSor: Combining Social and Semantic Web to Support the Analysis of Emergency Situations==
https://ceur-ws.org/Vol-747/paper4.pdf
SemSor: Combining Social and Semantic Web to Support
the Analysis of Emergency Situations
Philipp Heim Dennis Thom Thomas Ertl
Institute for Visualization and Institute for Visualization and Institute for Visualization and
Interactive Systems (VIS) Interactive Systems (VIS) Interactive Systems (VIS)
Universitätsstraße 38 Universitätsstraße 38 Universitätsstraße 38
Stuttgart, Germany Stuttgart, Germany Stuttgart, Germany
philipp.heim@vis.uni- dennis.thom@vis.uni- thomas.ertl@vis.uni-
stuttgart.de stuttgart.de stuttgart.de
ABSTRACT the situation itself is distributed in space or time so that one
In this paper we introduce SemSor, a system developed espe- single person alone has difficulties in getting an overview.
cially for the analysis of emergency situations. It constantly These cases hold true, e.g., in emergency operation centers
collects information from sources of the Social Web, maps (EOCs), where neither the agent who receives an emergency
it to unique resources in the Semantic Web and uses the call is onside nor the person who has made the call has
annotated information as basis for the situation analysis. If usually a comprehensive picture of the situation. However,
an emergency situation needs to get analyzed, four steps are having a comprehensive picture is especially important for
required: First, all information that is already known about the analysis of emergency situations in order to take the
this situation must be entered in the SemSor-GUI. Second, right actions and thus prevent all kinds of damage.
the entered information needs to be mapped to resources in To overcome the difficulties in analyzing emergency situa-
the Semantic Web. Third, using these resources as starting tions, we propose an approach that combines the advantages
nodes, a spreading activation is applied along the relation- of the Social Web with those of the Semantic Web. The idea
ships within the Semantic Web to find relevant Social Web is to scan Social Web entries, semantically annotate their
information. And fourth, the newly identified information is content and use spreading activation to find exactly those
visualized according to different dimensions and can be fil- entries that are useful for the analysis of a specific emergency
tered and explored by the user. In an iterative process, new situation. The idea of combining Social and Semantic Web
insights can be used to refine the query and thus improve to a Social Semantic Web has already been described e.g.
the activated information until a comprehensive analysis of in [10] and implemented in many applications, e.g. within
even complex situations is possible. the WeKnowIt-Project [6]. Also tools have been developed
that use this idea to support the analysis of emergency sit-
uations [8, 13]. In these tools the found Social Web entries
Categories and Subject Descriptors are often arranged on a map to provide an overview of the
H.3.1 [Content Analysis and Indexing]: Indexing meth- geographical extent, e.g. on the Interactive Fire Map [3], or
ods; H.3.3 [Information Search and Retrieval]: Infor- to extract relevant information via geographical filters [14].
mation filtering, query formulation, relevance feedback; H.5.2 Also popular are timelines that order Social Web entries ac-
[User Interfaces]: Graphical user interfaces (GUI) cording to the date of their creation and thus support an
understanding of the chronology of events [15]. However,
Keywords none of these approaches use spreading activation to find
semantically related Social Web entries automatically.
Social Web, Web 2.0, Semantic Web, Social Semantic Web, The general idea of spreading activation in semantic graphs
situation analysis, spreading activation, interactive informa- has first been introduced in [9]. Initially a set of starting
tion retrieval nodes is labeled with activation energy, which then is it-
eratively propagated to other nodes that are linked to the
1. INTRODUCTION starting nodes. Links can be weighted in order to control the
Analyzing emergency situations is difficult in cases where spreading of energy. This can be used, for instance, in rec-
either the agent who does the analysis is not close by and ommender systems to adapt the content of a web site to the
thus is not able to directly hear or see what is going on, or current context of its visitors. Next to the users’ concrete in-
formation needs, contextual information like location, time,
role, or weather conditions can be used to spread the activa-
tion differently and thus to find information that is relevant
with respect to a specific context [12]. In addition, every
user action can lead to refined link weights and activation
energies to account for individual preferences and interests.
However, this requires the semantic graph to be stored lo-
cally or on a server with write permissions in order to be able
to adapt the weights and activation energies accordingly.
.
Figure 1: SemSor architecture: Social Web entries are constantly crawled (A) and mapped to semantic
resources (B). If a situation has to be analyzed, all known information also needs to get mapped to semantic
resources (C), which function as starting nodes for the spreading activation (D). The found Social Web entries
are visualized and can be explored and filtered by the user (E). Gained insights, from the visualization or
from external sources (F), can then iteratively be used to improve the activated information.
In this paper we introduce an approach that applies spread- erences, are therefore mapped to unique semantic resources
ing activation in external semantic datasets and thus saves of datasets in the LOD cloud, e.g. DBpedia [7] or GeoN-
storage space and calculating capacities. Datasets in the ames [1] (Fig. 1, B). Thus the system automatically creates
LOD cloud [2] are accessed via SPARQL [5] queries to trig- a machine readable representation of the semantic that is
ger the spreading activation and thus to find semantically contained in the found Social Web entries, which can later
related resources. Besides the low system requirements, the on be used to support the analysis.
two main advantages of an outsourced spreading activation Once a certain situation needs to be analyzed (Fig. 1,
approach are: 1) The datasets are always up-to-date; no F), e.g. because of an incoming emergency call, everything
complicated methods for updating local copies are required. that is known about this situation must also be mapped
2) Semantic relationships of all kinds and domains are used to unique semantic resources (Fig. 1, C). These resources
to activate relevant information; spreading activation is not are then used as starting nodes for the spreading activation
restricted to a predefined set of resources, e.g. resources [9] that is applied to find all semantic resources that might
of a certain domain, but can include all domains contained be of relevance (Fig. 1, D). In a next step, all the Social
in the LOD cloud. With the SemSor system, we present an Web entries that have been annotated with at least one of
prototypical implementation of our approach that facilitates the activated resources are collected and form the result set,
the extraction of community information relevant to analyze which is presented to the user via multiple views (Fig. 1, E).
a certain emergency situation. Even though the spreading The result set can interactively be explored and filtered by
activation takes place externally, the user can rate the rel- the user in order to gain new insights about the situation.
evance of the found Social Web entries to refine the search Gained insights or news from external sources, e.g. from the
query and thus change the activation values until a thorough first responders, (Fig. 1, F) can then iteratively be used to
analysis can be achieved. refine the search query, thereby activate new resources in
In the following we first describe the general SemSor archi- the Semantic Web and thus improve the situation analysis.
tecture with all the components and steps that are required Due to the possibility to iteratively refine the search query,
for the analysis of emergency situations and provide further humans and computers can co-operate in this task.
details to each of the steps afterwards. This includes the
crawling and annotating of Social Web entries, the initial 3. CRAWLING AND ANNOTATING SOCIAL
query formulation, the spreading activation, the visualiza-
tion and filtering, and the interactive query refinement. At WEB ENTRIES
the end of the paper a conclusion and an outlook on future In order to use information that is contained in Social
work is given. Web entries to support the analysis of emergency situations,
the entries first have to be extracted and annotated by the
crawler component of SemSor. Even though the Social Web
2. SYSTEM ARCHITECTURE contains a huge amount of data, only a minimum of this
The SemSor System constantly scans Social Web sources, information needs to get stored in the SemSor database.
e.g. Twitter, Flickr and YouTube, for new entries (Fig. 1, In a first step, a broad multitude of Social Web data gets
A) and semantically annotates their textual content. Terms collected and evaluated according to a preconfigured met-
with a distinct meaning, e.g. geographical or temporal ref- ric that determines the a priori relevance of each individual
entry. Within this metric, different properties of an entry
like its source, the date and time of its creation as well as
location-based data get extracted and serve as basis to cal-
culate a weighted importance rating. The weights of the
metric can be configured according to the individual needs
of its users (e.g. a specific emergency response team) and
provide a basic means to decide, which entries should be
kept and which can be deleted if computational- or storage-
resources become short. Following the collection and a priori
evaluation, the Social Web entries are analyzed and certain
terms in their textual contents are automatically assigned to
unique resources in datasets in the LOD cloud by using ser-
vices like e.g. OpenCalais [4]. Once new entries have been
registered and evaluated, only their URLs and the URIs of
the assigned semantic resources have to be kept for subse-
quent steps.
4. INTERACTIVE SITUATION ANALYSIS
The SemSor system supports the whole situation analysis
process, including the initial query formulation, the search Figure 3: Search terms are interactively mapped
for relevant information via spreading activation, the visu- to resources in semantic datasets (A) that function
alization and filtering of the results as well as mechanisms as starting nodes for the spreading activation that
to iteratively refine the query. is applied along the links within these datasets (B
and C). As a result, semantically related resources
4.1 Initial Query Formulation get activated and thus Social Web entries annotated
The method of query is based on common question schemes with at least one of them get found (D).
of emergency calls according to relevant aspects of a situa-
tion: ”What has happened?”, ”Where did it happen?” and
”Who is involved?” (Fig. 2, B). The agent is supposed to pro- this facilitates finding information that is relevant for the
vide approximate answers to at least some of these questions analysis of a certain emergency situation.
and can further substantiate his query by providing bound- The spreading activation in SemSor is implemented mostly
aries for the temporal and spatial extent of the situation. as a remote process. On the client side, the process is only
In this process, the agent is assisted by an adaptive auto- triggered and controlled but is run completely within exter-
complete feature, which will try to interactively map given nal datasets on server side. Therefore SPARQL queries are
search terms to resources in the Semantic Web. Through- sent to the datasets to find resources related to the starting
out this process the definitions of proposed resources are nodes, which are then scored according to the semantical
provided in pop-up windows, e.g. corresponding Wikipedia and topological properties of their relationships.
articles (Fig. 2, A), to help users especially in the disam- Related resources are found based on an approach de-
biguation of ambiguous input terms. Based on this proce- scribed in [11]. Taking the starting nodes as roots, a breadth-
dure the system is able to get a reliable handle onto the first search (BFS) is applied to find all resources that are re-
relevant nodes in the Semantic Web. During an emergency lated to one of the starting nodes up to a predefined depth
situation, like e.g. the 2010 Haiti earthquake, disaster agents threshold. The depth threshold defines the number of nodes
can obtain a general overview of the situation by performing that are allowed between a starting node and a resource
a broad search on keywords like ”Earthquake” and ”Haiti”. that can be activated. Thus having e.g. a depth threshold
Based on the interactive mapping of search terms to seman- equal null restricts the activation radius to resources that are
tic resources the query is annotated by SemSor and con- directly connected to one of the starting nodes. For each re-
nected with resources in the Semantic Web. The nodes of lated resource that is found, the algorithm checks whether
this framework serve as the initially activated nodes in the Social Web entries have been crawled that are assigned to it
spreading activation procedure. (e.g. the Flickr entry is assigned to Fire in Fig. 3).
All those Social Web entries are then activated according
4.2 Spreading Activation to topological and semantical aspects and thereby scored.
Based on the initial activation of the user defined staring In our implementation the extent of activation depends on
nodes (Fig. 3, A), a homogeneous spreading activation is three aspects: 1) The length of the relationship, e.g. the
applied along the relationships between the resources in the length between the Flickr entry and the starting nodes is
datasets. In order to automatically activate semantically two, 2) the connection types within the relationships, e.g.
related resources that might also be relevant for the situ- ”district of” or ”capital”, and 3) the classes of the interme-
ation analysis, the activation happens along the instance- diate resources, e.g. ”Port-au-Prince” is a city. It is also
relationship layer (Fig. 3, B) as well as the class-relationship possible to define certain connection types or classes that
layer (Fig. 3, C). Thus the resulting set of Social Web en- should not be used to spread the activation, which is use-
tries is not limited to those containing references to one of ful if someone is not interested in relationships that contain
the user defined starting nodes only, but also includes entries certain instances or connections.
referring to resources that are within a distinct semantic ra-
dius around the user defined starting nodes (Fig. 3, D);
Figure 2: SemSor GUI: Search terms are interactively mapped to semantic resources (A) that together form
the search query (B). Relevant information from Social Web sources is found automatically and shown in
a list (C). Single entries can be examined in detail (D) and filters can be formulated according to various
dimensions, e.g. spatial filters (E) or temporal filters (G).
4.3 Visualization and Filtering and by dragging additional tags from the items to the query
Based on their activation, the Social Web entries are visu- fields. High ratings of some Social Web entries can then lead
alized in the SemSor GUI. While the search still continues, to the activation of semantic resources which are directly
all Social Web entries that were already discovered by the connected to those entries (Fig. 4, D). Thus after rating the
activation are presented through the result browser in dif- items, the agent is given the possibility to restart his search,
ferent user-selectable views (e.g. tabular, map and statistic but this time, the spreading activation will execute starting
view). Every view provides the opportunity to obtain pic- also from the newly activated items (Fig. 4, E). Through this
tures, videos and other user generated content related to the ”pollination” and subsequent activation of remote nodes in
situation (Fig. 2, D). The standard view is a listing of en- the semantic graph the agent is given the chance to discover
tries sorted by their individual semantic relevance (Fig. 2, relevant regions and new Social Web entries that were not
C). In order to get an overview of the spatial distribution included or even near his initial query (Fig. 4, F). Based on
of possibly relevant entries, the agent can view them on a the filters and the interaction procedure, the agent is able to
map that can also be used to formulate geographical con- cope with the enormous flood of Social Web data that can
straints, e.g. to show only results that refer to a certain be found in connection with emergency situations and make
geographic region (Fig. 2, E). To further explore the set beneficial use of them.
of results a time line offers an overview over the temporal
distribution of events (Fig. 2, G) and allows to formulate 5. CONCLUSIONS AND FUTURE WORK
temporal constraints, e.g. to show only results that refer to
a specific period of time. To further explore the quality and In this paper, we described an approach that combines the
diversity of results the agent has the opportunity to analyze Social and the Semantic Web in order to support the analysis
diagrams that show the composition of chosen result-subsets of emergency situations. Exploiting information from the
by author, location or tag-categories. By acquiring these ini- Social Web is especially useful when: (1) the situation is
tial impressions, the agent can further asses the nature and distributed in space or time, (2) first responders are not on
extent of the situation and initiate subsequent steps. site, or (3) the situation cannot or only partly be observed;
e.g. this can be the case if a situation takes place within a
building or is hidden behind some obstacles.
4.4 Interactive Query Refinement The automatic semantic annotation of entries in the So-
At this point the advantages of the interactive features cial Web as well as the interactive mapping of entered search
in SemSor come into play. If the resulting set generated terms to unique resources in the Semantic Web allows infor-
by the initial spreading activation (Fig. 4, A and B) and mation to be found not only by string matching but also
diminished by the user defined filters is yet not sufficient, according to its meaning. Therefore spreading activation is
the agent can further refine and expand his initial query by applied in external semantic datasets that offer both up-to-
rating single result items on a continuous scale (Fig. 4, C) date and comprehensive information on all kinds of topics.
vent a possible emergency situation from happening.
6. ACKNOWLEDGMENTS
The prototypical implementation of our approach, the
SemSor system, was developed within a student project by
Clint Banzhaf, David Schmid, Dominik Jäckle, Edwin Pütt-
mann, Jochen Seitz, Johannes Dilli, Marijo Macet, Nico
Ploner, Steffen Bold, Stephan Engelhardt and Thomas Mich-
elbach. We would like to thank them for their excellent work.
Furthermore we thank Bernhard Schmitz, Michael Wörner
and Thomas Schlegel for co-supervising the project.
7. REFERENCES
[1] Geonames. http://www.geonames.org/, 2006.
[2] Linking open data community project (lod).
http://esw.w3.org/SweoIG/TaskForces/
CommunityProjects/LinkingOpenData, 2007.
Figure 4: Once an entry has been found via spread- [3] San diego wildfires 2007 interactive fire map.
ing activation (A and B), the agent can rate it as http://www.signonsandiego.com/firemap/, 2007.
relevant (C) and thus indirectly expand the search [4] Opencalais. http://www.opencalais.com/, 2008.
query (D). This leads to the activation of other [5] Sparql protocol and rdf query language.
nodes (E) and thus can produce Social Web entries http://www.w3.org/TR/rdf-sparql-query/, 2008.
as result set that are only distantly related to the [6] Weknowit project. http://www.weknowit.eu, 2010.
user defined search query (F). [7] S. Auer, C. Bizer, G. Kobilarov, J. Lehmann,
R. Cyganiak, and Z. Ives. DBpedia: a nucleus for a
web of open data. In Proc. of the 6th Int. Semantic
Because of the outsourced spreading activation and the Web Conf. (ISWC’08), volume 4825 of LNCS, pages
minimum information that is needed to store the annotated 722–735. Springer, 2008.
Social Web entries – an entry is represented by its URL and [8] J. Borsje, L. Levering, and F. Frasincar. Hermes: a
the URIs of the annotated resources only – SemSor is able semantic web-based news decision support system. In
to handle the huge amounts of available data and find in- Proc. of the 2008 ACM symposium on Applied
formation relevant for the analysis of a certain emergency computing, SAC ’08, pages 2415–2420, New York, NY,
situation. The found information is visualized in multiple USA, 2008. ACM.
views and can be explored and filtered by the agent based [9] A. M. Collins and E. F. Loftus. A spreading-activation
on individual information needs. If the result set is not yet theory of semantic processing. Psychological Review,
sufficient, the query can interactively be expanded or nar- 82(6):407 – 428, 1975.
rowed down. Single entries can be rated as relevant or as [10] T. Gruber. Collective knowledge systems: Where the
irrelevant which changes the starting nodes and thus can re- social web meets the semantic web. Web Semantics:
sult in other entries to be found by the spreading activation. Science, Services and Agents on the World Wide Web,
The query can be refined until a sufficient analysis of the 6:4–13, 2008.
emergency situation is possible.
[11] P. Heim, S. Hellmann, J. Lehmann, S. Lohmann, and
In its current implementation, the SemSor system is most
T. Stegemann. Relfinder: Revealing relationships in
suitable to analyze current or past emergency situations (cp.
rdf knowledge bases. In Proc. of the 4th Int. Conf. on
Fig. 2). Since many people use the Social Web to comment
Semantic and Digital Media Technologies (SAMT
on emergency situations, relevant information is available
2009), pages 182–187. Springer, 2009.
even while a situation is happening. Together with the fact
[12] T. Hussein and J. Ziegler. Adapting web sites by
that only seconds are required from the time a new com-
spreading activation in ontologies. In Proc. of the Int.
ment is uploaded to when it can be found in SemSor, it is
Workshop on Recommendation and Collaboration
already possible to facilitate the analysis of current and past
(ReColl ’08), 2008.
emergency situations.
However, analyzing current or past emergency situations [13] N. Ireson. Local community situational awareness
often cannot prevent them from happening. In order to pre- during an emergency. In Proc. of the IEEE Int. Conf.
vent emergency situations a preventive analysis is required. on Digital Ecosystems and Technologies (DEST 2009),
First signs of a forthcoming emergency situation need to be pages 49–54. IEEE, 2009.
detected and interpreted in the right way so that the right [14] C. P. Julio and C. A. Iglesias Fernández. Disasters 2.0:
actions can be initiated. Besides the agent triggered search, Application of web 2.0 technologies in emergency
this would require SemSor to automatically scan the Social situations. In Proc. of the 6th Int. ISCRAM Conf.,
Web entries for new topics and trends and iteratively pro- Gothenburg, Sweden, 2009.
duce an overview of the current situation. If certain topics [15] S. B. Liu and L. Palen. Spatiotemporal mashups: A
get popular or unusual changes can be detected an alarm survey of current tools to inform next generation crisis
could be raced automatically that could force an agent to support. In Proc. of the 6th Int. ISCRAM Conf.,
check the situation and decide on the right actions to pre- Gothenburg, Sweden, 2009.