We present WiBi Explorer, a new Web application developed in our laboratory for visualizing and exploring the bitaxonomy of Wikipedia, that is, a taxonomy over Wikipedia articles aligned to a taxonomy over Wikipedia categories. The application also enables users to explore and convert the taxonomic information into RDF format. The system is publicly accessible at wibitaxonomy.org and all the data is freely downloadable and released under a CC BY-NC-SA 3.0 license.
Knowledge modeling is a long-standing problem which has been addressed in a
variety of ways (see [
In neither of these resource types, however, is any explicit attention paid to
hypernymy as a distinct relation type. Instead, hypernymy has been proven to
be a relevant relation type capable of ameliorating systems in several hard tasks
in Natural Language Processing [
WiBi [
The bitaxonomy is built thanks to a three-phase approach that i) rst builds
a taxonomy for the Wikipedia pages, then ii) leverages this partial information
to iteratively infer new hypernymy relations over Wikipedia categories while at
the same time increasing the page taxonomy, and nally iii) re nes the obtained
category taxonomy by means of three ad-hoc heuristics that cope with structural
problems a ecting some categories. As a result, a bitaxonomy is obtained where
each element - either page or category - is associated with one or more hypernyms
and where elements of one taxonomy are aligned (i.e, linked) to elements of the
other taxonomy. In order to transfer hypernymy knowledge from either one of
the two Wikipedia sides to the other side, the whole process remarkably, and as
a key feature, exploits categorization edges (here called cross-edges ) manually
provided by Wikipedians, which connect any page on one side to its categories
on the other side and vice versa. Extensive comparison has been carried out
on two datasets of 1,000 pages and categories each, against all the available
knowledge resources, including MENTA, DBpedia, YAGO, WikiTaxonomy and
WikiNet (for an extensive survey, see [
Here we present a Web-based visual explorer for displaying the two aligned taxonomies of WiBi, centered on any given Wikipedia item of interest chosen by the user. The interface easily integrates search facilities with customization tools which personalize the experience from a user's point of view. The home page. An excerpt of the interface's home page is shown in Fig. 1(a). As can be seen, this page has been kept very clean with as few elements as possible. On the top of the page a navigation bar contains links to i) the about page, which contains release information about the website content, ii) a download area, where it is possible to obtain the data underlying the interface and iii) the search page, which represents the core contribution of this work.
The search page mainly contains a text area in which the user is requested to input her query of interest, additionally opting for searching through either the page inventory, the category inventory or both, thanks to dedicated radio buttons. After the query is sent, the search engine tries to match the input text against the whole database of Wikipedia pages (or categories) and, if a match is found, the engine displays the nal result to the user. Otherwise, the query is interpreted as a lemma and the user is returned with the (possible) list of all Wikipedia pages/categories whose lemma matches against the query. The result page. Starting from the Wikipedia element provided by the user, the objective of the result page is to show a relevant excerpt of the bitaxonomy, that is, the nearest (or more relevant) nodes connected to it, drawn from both of the two taxonomies. To do this, WiBi Explorer performs a series of steps: 1. Start a DFS of maximum length 1 from the given element p of a taxonomy. As a result, a subgraph ST1 = (SV1; SE1) is obtained; (a) WiBi Explorer's home page.
(b) Result for the ISWC Wikipedia page. 2. Collect all the nodes (p) belonging to the other taxonomy (i.e, those whose crossedges are incident to p). Start a DFS of maximum length 2 from each element in (p). As a result, a subgraph ST2 = (SV2; SE2) is obtained; 3. Display ST1 and ST2, as well as all the possible cross-edges linking nodes of the two subgraphs. Prune out low-connected nodes from the displayed bitaxonomy.
As a result, the interface displays a meaningful excerpt of the two taxonomies, centered on the issued query. The result for the Wikipedia page International Semantic Web Conference is shown in Fig. 1(b).
Customization of the view Since a user might be interested in a more general view of the bitaxonomy, two additional sliders are provided to the user in order to manually adjust the two maximum depths 1 and 2 (see Fig. 1(b) on top). Moreover, the interface provides the user with the capability to click on nodes and interactively explore di erent parts of the taxonomy. The application thus acts as a dynamic explorer that enables users to navigate through the structure of the bitaxonomy and discover new relations as the visit proceeds. 4
Interestingly, data can also be exported in RDF format, in line with recent
work on (linguistic) linked open data and the Semantic Web [
We have proposed the Wikipedia Bitaxonomy Explorer, a new, exible and
extensible Web interface that allows the navigation of the recently created Wikipedia
Bitaxonomy [
The authors gratefully acknowledge the support of the
ERC Starting Grant MultiJEDI No. 259234.
The authors also acknowledge support from the LIDER project (No. 610782), a Coordination and Support Action funded by the EC under FP7.