This paper introduces TEWS |Twitter Events on the Semantic Web, pronounced like \news"|a semantic web tool for detection and representation of events taking as an input the social stream Twitter. The tool assists the user throughout a complete processing chain, starting from the detection of events on Twitter, their modeling and representation following the semantic web principles, to their storing in an RDF knowledge base that can be further published on the Web of Data.
One of the goals of the Linked Open Data (LOD) initiative is to structure and
interconnect data on the web by using semantic web technologies, such as the
Resource Description Framework (RDF), thus taking the web of today up to
the level of a veritable information network [
The originality of the approach lies in the use of tweets produced by
conventional media (news agencies) in order to guide the event detection process on
the tweets produced by ordinary users (the crowd). Contrary to previous works,
in which the focus falls on event detection on Twitter [
TEWS takes as an input a corpus of tweets. We have developed a protocol for the continuous constitution of a corpus by the help of the Twitter4J3 API, which is designed to collect tweets from the social stream. As shown on the left hand side of Fig. 1 (steps 1{3), we are interested in two sources of information: (1) the conventional media, seen as a reliable and uninterrupted source of information about events (such as ABC News, Al Jazeera, CNN, BBC, etc.), and (2) the crowd (ordinary users). The latter is ltered by considering only tweets that contain a hashtag or a keyword (a location) from the media tweets corpus. The resulting set of tweets is continuously fed to TEWS in real time.
The overall events detection work ow and their RDF representation is given
in Fig. 1. Steps 1{3 represent the corpus extraction procedure discussed above.
TEWS is responsible for the actual extraction of events and their modeling and
storing as RDF triples, given in steps 4{6. Step 7 illustrates the option that these
data are further included to the web of data following the well-established
publication protocol. As a design choice, the tool makes use of the MABED system
[
Event Detection. MABED is based on statistical methods for the identi
cation of important events discussed by the users in a tweets corpus. In the rst
place, an event is de ned by a salient topic identi ed by a sudden increase of the
occurrence of certain key-words. An event is de ned by a main key-word and a
set of related terms. The e ciency of the approach is demonstrated in [
Event Modeling. After comparing various existing ontologies, we have chosen
the WikiTimes model for representing events, for it appeared to be the closest to
our vision of an event. WikiTimes allows for the creation of two types of resources
{ an event and a story (a sequence of events). We focus on simple events only.
3 http://twitter4j.org/en/index.html
Each event is anchored in the time and is characterized by several properties,
described in a speci c vocabulary, such as place and date of occurrence and
involved entities, as well as additional information (Fig. 2).
We explain how TEWS works through a case study. A video illustrating this
scenario is made available online4. The initial page of TEWS (Fig. 3(a)) allows
to de ne all the parameters our tool needs. It has three main areas: i) Corpus
selection: in this box we can specify the date (year, month, day) related to the
tweets collection. The tool will retrieve only tweets published in the speci ed
period. ii) Detection Parameters : this area allows to introduce the parameters
related to the MABED system [
Fig. 3(b) shows the results related to the speci ed period and the
parameters introduced in the previous step. We can observe that TEWS detects ten
events. Each event is described by the date, the location, the involved entities,
the sources and the textual description. From the result list we can select one
(or more) lines and visualize the corresponding RDF triples automatically
generated by TEWS. In the example, we select the rst event related to the Yemen
earthquake and the tool produces the corresponding RDF knowledge. The RDF
triples are visualized at the bottom of the page. The WikiTimes [
We introduced TEWS |a new Twitter event detection tool that produces RDF triples. The tool exploits conventional media sources to extract important keywords to trace phenomena on the stream of tweets. TEWS supplies a complete processing chain that combines state-of-the-art Twitter event detection with semantic web technologies to produce new knowledge to share on the web of data. As future work, we plan to improve the natural language processing aspects of the tool (such as named entity recognition and word sens disambiguation in order to lter out false entities, (i.e., exploit DBPedia). We also plan to integrate other sources of information to enrich the knowledge extraction from Twitter supplying a more complete description of the detected events.