Content uploaded by Gregory Grefenstette
Author content
All content in this area was uploaded by Gregory Grefenstette on Nov 13, 2015
Content may be subject to copyright.
Content uploaded by Gregory Grefenstette
Author content
All content in this area was uploaded by Gregory Grefenstette on Nov 13, 2015
Content may be subject to copyright.
Transforming Wikipedia into a Search Engine for Local
Experts
Gregory Grefenstette
Inria Saclay/TAO, Rue Noetzlin - Bât 660
91190 Gif sur Yvette, France
gregory.grefenstette@inria.fr
Karima Rafes
Inria Saclay/TAO, Rue Noetzlin - Bât 660
91190 Gif sur Yvette, France
Karima.rafes@inria.fr
ABSTRACT
Finding experts for a given problem is recognized as a
difficult task. Even when a taxonomy of subject expertise
exists, and is associated with a group of experts, it can be
hard to exploit by users who have not internalized the
taxonomy. Here we present a method for both attaching
experts to a domain ontology, and hiding this fact from the
end user looking for an expert. By linking Wikipedia to
this same pivot ontology, we describe how a user can
browse Wikipedia, as they normally do to search for
information, and use this browsing behavior to find
experts. Experts are characterized by their textual
productions (webpages, publications, reports), and these
textual productions are attached to concepts in the pivot
ontology. When the user finds the Wikipedia page
characterizing their need, a list of experts is displayed. In
this way we transform Wikipedia into a search engine for
experts.
1. Introduction
In large organizations, such as multinational corporations,
universities, and even large research centers, it can be difficult
to know who is an expert about a given subject. A common
response to this problem is to create an ontology of expertise
and manually or automatically assign experts labels from this
ontology. Beyond the cost or effort needed to produce the
ontology, this solution creates an additional problem. Once
such a knowledge base of experts exists, the searcher still has
to internalize the ontology labels and their meaning in order to
find the expert. The difficulty the user faces explains why
some expert knowledge bases are found useful for one division
in a large organization but be useless for another division
which does not share the same terminology or perspective
(Hahn & Subrami, 2000). We propose a method for finding
experts that hides the pivot ontology from the user, and allows
the searcher to browse Wikipedia, a resource that he or she is
probably familiar with, in order to find his or her local expert.
2. Solution
Our solution involves mapping experts from a givein
organization onto a domain ontology for their expertise, and
then mapping Wikipedia articles into the same domain
ontology. To find an expert on a certain field, the user searches
the subject of interest in Wikipedia. A tab added to the
Wikipedia page can be clicked to reveal a list of experts
concerning the some topic mentioned in the page.
2.1 Example
Before explaining the mechanisms and language resources
involved, let us see an example. In this example, our local
experts are any of the research teams in the French nation-wide
computer science public research institute, Inria1. The Inria
Institute employs 3600 scientists spread over 200 research
teams, each specializing in some branch of computer science
and mathematics. In our example, finding an expert will mean
finding a team who is competent to answer questions about a
given subject.
Figure 1 An extra tab appears while browsing Wikipedia once the expert finder
module is activated by a logged in user. Here the tab is labeled 'Inria' since we
are searching for experts inside the Inria Institute.
Figure 2. Clicking on the tab expands a box that shows the Inria teams that are
associated with subjects on the Wikipedia page. For example, one subject
mentioned on the page is “Gaussian processes” and 3 Inria teams that work in
this domain are listed ASPI, ATHENA, and BIGS, with their expanded team
names. Clicking on “Gaussian processes” goes to an ACM 2012 ontology page
for this concept. Clicking on a team name goes to a web page from their 2014
annual report where a project involves “Gaussian processes”. On this page, the
user can find team members who are experts in the area. (See Figures 3 and 4)
In our solution, when someone is looking for an expert in a
given subject inside the Inria institute, an additional tab
appears to the user on the Wikipedia interface2. In Figure 1,
1https://en.wikipedia.org/wiki/Inria
2 This tab appears when the user is logged in, and has the
resource module for expert finding, this tab appears while
the user browses http://en.wikipedia.org. Wiki resource
2 This tab appears when the user is logged in, and has the
resource module for expert finding, this tab appears while
the tab appears with the label “Inria” to the right of the
“Article” and “Talk” tabs above the article title. Clicking on
the tab expands a box listing the ACM subjects found in the
articles and the Inria research teams treating those subjects.
Both subjects and teams (see Figure 2) are linked to pages
outside Wikipedia, so that Wikipedia has become a search
engine with the user browsing towards their query (here
“Autoregressive Models”, with the pull down expert box
corresponding the search engine results page, leading to
outside content. The user can find the Wikipedia article closest
to his or her concern, and use the expert finding tab to find
local experts who know about the subjects on the page. This
seems to us a natural and intuitive method for finding experts
that obviates the need for learning the ontology by which the
experts are connected to the topic page. Even if the connecting
ontology terms are explicitly displayed in the results, the user
need not ever use them in an explicit query.
Figure 3. The Inria team annual report page found by following the link in the
expert finding module. There the user sees that “Valerie Moribet” is involved in
a project using Gaussian processes, and would probably be a good expert contact
for finding out about Auto-regressive models.
Figure 4. The ACM 2012 is the latest version of a computer science ontology
created by the Association for Computing Machinery. "Gaussian processes" is a
hyponym of “Theory of Computation”. This specific concepts provides a link
between the Wikipedia web page on “Autoregressive models” and the Inria team
experts, but the searcher need not understand the ACM hierarchy nor know its
contents in order to make the connection.
2.2 Underlying Language Resources
The connections between Wikipedia content and local experts
pass through a shared pivot ontology. Both Wikipedia page
content and the expert profiles are mapped into the same
ontology categories which provide a pivot, or link between
them. In our implemented example, we used the ACM 2012
Classification schema3 as the shared ontological space. Here is
an entry in this ontology:
<skos:Concept rdf:about="#10010296" xml:lang="en">
the user browses http://en.wikipedia.org. Wiki resource
modules are additions that anyone can develop and activate.
https://www.mediawiki.org/wiki/ResourceLoader/Developin
g_with_ResourceLoader explains them further.
3 http://www.acm.org/about/class/class/2012
<skos:prefLabel xml:lang="en">Gaussian processes</skos:prefLabel>
<skos:altLabel xml:lang="en">gaussian process</skos:altLabel>
<skos:inScheme rdf:resource="http://totem.semedica.com/taxonomy/The ACM
Computing Classification System (CCS)"/>
<skos:broader rdf:resource="#10010075"/>
This entry gives a synonym for “Gaussian processes”, an
internal ACM code for this concept (10010296), and a link to a
hypernym (10010075), “Theory of computation”. This SKOS
taxonomy, augmented by any Wikipedia redirects as additional
synonyms, was converted into a natural language processing
program that associates the internal ACM code to raw English
text (delivered as resource with this paper).
Wikipedia text was extracted from a recent dump.4 The text
was extracted from each article, tokenized, lowercased, and
transformed into sentences. Each sentence was passed through
NLP program associating ACM codes to each article. Any
ACM concept appearing in more than 10,000 articles was
eliminated as too general5.
The source for expert profiles in our example were the public
web pages of Inria teams 2014 activity reports6. Each page
was downloaded, boilerplate removed, text extracted,
tokenized lowercased and split into sentences, as with the
Wikipedia text.
In all 3123 Inria web pages were associated 129,499
Wikipedia articles were tagged one or more of with 1049
different ACM codes.
2.3 Variations
Instead of using annual reports to create expert profiles, one
could instead use the publications of researchers from a given
team or research center.
In place the ACM hierarchy, one could extract a taxonomy of
Wikipedia categories and subcategories7.
For example, one could use MeSH as the anchor ontology and
publications of doctors at local hospital to transform Wikipedia
into a search engine of specialists for medical problems.
3. Related Work
Demartini (2007) proposed using Wikipedia to find experts by
extracting expertise from Wikipedia describing people, or from
Wiki editors that edited page corresponding to a given topic.
West et al. (2012) tried to characterize what makes Wikipedia
experts. Our approach allows us to connect to people who do
not appear in Wikipedia either as subjects or editors.
4. CONCLUSION
We have presented a system for finding local experts using
Wikipedia. It is constructed using a pivot ontology, indexing
Wikipedia pages, and some textual representation of the
experts (web pages, reports, or publications). The pivot
ontology must be represented as a natural language processing
resource, i.e. a resource that a program can apply to natural
language text, that is applied to both Wikipedia pages and the
textual representation of experts, pushing them into the same
space. Once this mapping is done, a Wikimedia resource
loader dynamically makes the connection once the user click
4 http://dumps.wikimedia.org/biwiki/latest
5 For example, ACM has a concept ‘Women’ under ‘Gender’.
‘Women’ was eliminated while ‘Gender’ was retained.
6 https://raweb.inria.fr/rapportsactivite/RA2014/index.html
7 Starting from, for example,
https://en.wikipedia.org/wiki/Category:Computer_science
on the expert fining tab, so neither the text of Wikipedia, not
the expert profile text need be altered. An additional advantage
of this system is that the user seeking an expert does not
interact explicitly with the pivot ontology, but only with
Wikipedia and the original textual representations of the
experts. As the general public becomes more used to using
Wikipedia to find information, they are able to find the best
page that characterizes their need. And Wikipedia is
transformed into a local expert search engine.
5. References
Demartini, Gianluca. "Finding Experts Using Wikipedia."
FEWS 290: 33-41. 2007.
Hristoskova, Anna, Elena Tsiporkova, Tom Tourw, Simon
Buelens, Mattias Putman, and Filip De Turck. "Identifying
experts through a framework for knowledge extraction from
public online sources." In 12th Dutch-Belgian Information
Retrieval Workshop (DIR 2012), Ghent, Belgium, pp. 19-22.
2012.
Hahn, Jungpil, and Mani R. Subramani. "A framework of
knowledge management systems: issues and challenges for
theory and practice." In Proceedings of the twenty first
international conference on Information systems, pp. 302-312.
Association for Information Systems, 2000.
West, Robert, Ingmar Weber, and Carlos Castillo. "A data-
driven sketch of Wikipedia editors." In Proceedings of the 21st
international conference companion on World Wide Web, pp.
631-632. ACM, 2012.
