Ontology-driven question answering system with semantic web services support

Abstract

Nowadays the internet is becoming a huge dump of documents, links and all other sorts of information. Most common possibilities to explore this information are information retrieval applications such as web search engines. Despite the fact that search engines are doing an excellent job, they still return too much inaccurate information. The solution to this problem can be found in the form of question answering systems, where the user gives a question in natural language, similarly to talking with another person. The answer is the exact information instead of a list of possible results. This paper presents the design of our ontology-driven question answering system with semantic web services support.

Full text

Abstract—
Nowadays the internet is becoming a huge dump of
documents, links and all other sorts of information. Most common
possibilities to explore this information are information retrieval
applications such as web search engines. Despite the fact that search
engines are doing an excellent job, they still return too much
inaccurate information. The solution to this problem can be found in
the form of question answering systems, where the user gives a
question in natural language, similarly to talking with another person.
The answer is the exact information instead of a list of possible
results. This paper presents the design of our ontology-driven
question answering system with semantic web services support.
Keywords—Ontology, Question answering system,
Semantic web, Web services
.
I. INTRODUCTION
N today’s world the majority of information is accessible via
the World Wide Web. A common way to access this
information is through information retrieval applications like
web search engines. We already know that web search engines
flood their users with enormous amount of data from which
they cannot figure out the essential and most important
information.
These disadvantages can be reduced with question
answering systems. The basic idea of question answering
systems is to be able to provide answers to specific question
written in natural language. The main goal of question
answering systems is to find a specific answer. The answers
can be retrieved from domain-specific knowledge corpuses or
other external resources like web services.
This article is segmented into eight chapters. The following
chapter describes ontologies and semantic description of
domain-specific knowledge. The third chapter describes the
process used for ontology mapping to the relational database.
Manuscript received October 26, 2010.
Borut Gorenjak is with the Faculty of Electrical Engineering and
Computer Science, University of Maribor, Smetanova ulica 17, 2000,
Maribor, Slovenia. (phone: +386-2-220-7460; fax: +386-2-220-7272; e-mail:
borut.gorenjak@uni-mb.si).
Marko Ferme is with the Faculty of Electrical Engineering and Computer
Science, University of Maribor, Smetanova ulica 17, 2000, Maribor,
Slovenia. (phone: +386-2-220-7253; fax: +386-2-220-7272; e-mail:
marko.ferme@uni-mb.si).
Milan Ojsteršek is a teaching professor of the Faculty of Electrical
Engineering and Computer Science, University of Maribor, Smetanova ulica
17, 2000, Maribor, Slovenia. (phone: +386-2-220-7451; fax: +386-2-220-
7272; e-mail: ojstersek@uni-mb.si).
The following chapter interprets the use of question templates.
The fifth chapter describes integration of external knowledge
resources. This chapter also explains semantic description of
web services. The sixth chapter reveals the importance of user
behavior. The seventh chapter describes the architecture and
processes of our question answering system. Chapter eight
concludes with the summary and suggestions for our future
work.
II. SEMANTIC DESCRIPTION OF DOMAIN SPECIFIC
KNOWLEDGE
The majority of information available on the web is suitable
for human use. This is the main reason why computer
applications have a problem understanding of this data [2].
Fortunately, this problem can be solved by using the
semantic web. Semantic web is an extension of the World
Wide Web. As the name itself suggests, the purpose of the
semantic web is to precisely define unambiguous computer
understandable metadata, thus enabling computers and people
to work in cooperation [4]. One of the most important
components of the semantic web are ontologies which can
significantly enhance understanding and description of
information.
Ontology is a formal representation of knowledge as a set of
concepts within a specific domain, and the relationships
between those concepts. It is used to reason about the entities
within that domain, and may be used to describe the domain.
Contemporary ontologies share many structural similarities,
regardless of the language in which they are expressed. Most
ontologies describe individuals (instances), classes (concepts),
attributes, and relations [1].
Web Ontology Language (OWL) is a formal knowledge
representation language for authoring ontologies. Ontology
languages allow users to write explicit, formal
conceptualizations of domains models. The main OWL
requirements are: a well-defined syntax, well-defined
semantics, efficient reasoning support, sufficient expressive
power and convenience of expression. OWL is built on top of
RDF and RDF Schema, and uses RDF's XML syntax. OWL
was designed to be interpreted by computers and not for being
read by people.
RDF Query Language (SPARQL) is a SQL-like language
for querying RDF data.
Ontology-driven question answering system
with semantic web services support
Borut Gorenjak, Marko Ferme, Milan Ojsteršek
I
Advances in Communications, Computers, Systems, Circuits and Devices
ISBN: 978-960-474-250-9
199

III. ONTOLOGY MAPPING TO THE RELATIONAL DATABASE
As we describe in the previous chapter, we needed a way to
formalize our ontologies. At a beginning we chose OWL for
our knowledge presentation. We have been using Protégé, a
free, open source ontology editor and knowledge-base
framework [3]. It is a great tool for creating semantic web
content, but we were concerned with its suitability for our end
users. Our users are ordinary people, who do not know
anything about ontologies and semantic web. OWL contains
much more than we needed for our system. We also needed
support for phrases and their synonyms.
All this led us to the conclusion that we have to build our
own ontology representation. We took the idea of OWL, which
has been reduced with irrelevant elements. We added Domain,
Process, Phrases and Synonyms to our solution. We also added
the semantic description of methods and parameters, which
will be described in detail in the fifth chapter. Our ontology
mapping to relational database is shown in the fig.1.
It is important that the system is able to provide answers as
fast as is possible. This is the main reason, why we build the
whole idea as a relational database solution. We believe that
relational database is an optimal solution for us in terms of
speed of data searching.
PROCESS
PROPERTY INSTANCES
CLASS INSTANCES
DOMAIN
METHOD
PARAMETER
CLASS
PROPERTY PROPERTY PROPERTY
SYNONYM
PHRASE
Fig. 1 Ontology mapping to relational database
IV. USING QUESTION TEMPLATES
Natural language processing is a domain of Computer
Linguistics. Programs and algorithms should behave like they
understand natural language [5]. Natural language is
ambiguous and contains many synonyms, which can be
understood differently, depending on the context of the
sentence or even paragraph. The key to understand the
importance of the sentence is identification of entities.
Methods for determining of the meanings of phrases are
generally based on the use of a large knowledge corpus. Most
of those methods are slow, since they use a large amount of
data and the results are average. This applies to the Slovenian
language since there is currently no good enough semantic
dictionary for it. Therefore, we used a completely different
approach and introduced the question templates in the context
of domain-specific knowledge.
Question templates are a bridge between sentences and
ontology. They are used as a mapping between relations and
objects. Templates can be equated with the ontology as formal
presentation skills in the context of a domain. Elements of the
question templates are entities composed of phrases,
synonyms, class properties or even method parameters. We
have basic and complex templates. Basic templates are
composed of a question template that is related to a single
answer template. In complex cases, the question templates can
be related to the template of the second question (sub-
questions) if the user didn’t provide enough information for a
unique response. All this leads us to question answering
dialog.
Example of basic question template:
What is the e-mail address of [Person_Name]
[Person_Surname]?
What | is | the | of – words
e-mail address – phrase
[Person_Name] – ontology driven data that represent class
Person and its property Name
[Person_Surname] – ontology driven data that represent
class Person and its property Surname
Question template above has only one answer template:
[Person_Name] [Person_Surname] e-mail address is
[Person_Email].
e-mail address – phrase
[Person_Name] – ontology driven data that represent class
Person and its property Name
[Person_Surname] – ontology driven data that represent
class Person and its property Surname
[Person_Email] – ontology driven data that represent class
Person and its property Email
Advances in Communications, Computers, Systems, Circuits and Devices
ISBN: 978-960-474-250-9
200

V. EXTERNAL KNOWLEDGE RESOURCES
As our system was developed as an applicative project, we
had a special request. All information that our question
answering system can handle, cannot be presented as an
ontology. Certain information must nevertheless be calculated
and that means we have to obtain that certain data from
external source.
The most logical approach was to use web services. Web
services are typically application programming interfaces or
Web APIs that are accessed via Hypertext Transfer Protocol
(HTTP) and executed on a remote system hosting the
requested services.
Because we couldn’t get all the information as web services,
we had to extend that process to other ways of calling
methods. At the end we added the ability to call local DLL
libraries and stored procedures from the database. As all three
ways require method calls with the parameters, we need to
provide a way to describe the methods and parameters. We
consider it ideally to describe them with semantics. So we
expanded our ontology representation with method and
parameter description.
But of course we also had to develop a special software
wrapper that could figure out from the semantic description
how to call methods with certain values for parameters and
how to return and transform the returned calculated values.
VI. USER TRACKING
In applications that run as web applications we don’t have
any control about user inputs. So it is wise to track all user
activities during the use of application.
Our system offers active and passive user tracking. Active
tracking allows users to tell their opinions about certain answer
with simple button click. The system even allows users to
comment replied answers. Passive tracking has a whole
different approach which is very unobtrusive and users don’t
even know about it. Passive tracking uses client-side cookies
for anonymous user tracking. We are trying to detect a context
switch, which tells us about user’s satisfaction.
With such actions we can constantly update our knowledge
database. We can even track user questions and answers
returned by our system. If we encounter a certain amount of
errors in the responses, we can take appropriate action such as
template rebuilding or restructuring of semantic data
representation.
On the other hand we can generate all kind of statistics that
helps us understand our users’ behavior.
VII. QUESTION ANSWERING PROCESS
Question answering process always starts with the user’s
input. The whole question answering process is shown in Fig.
2. When the user enters a sentence, it fires the process for
detecting entity candidates. We already know that templates
are composed of entities and that’s why we have to find the
appropriate candidates for template matching process. Entity
candidate detection process uses our domain specific
knowledge database for detecting entities. Entities are
recorded as instances of our ontology. At that stage our
process uses a dialog states table, where the actual state of user
dialog is stored. That is very important for understanding what
data has already been entered by the user.
The most important task in the whole process is the template
matching process. This process must decide which question
template is the most similar to the sentence entered by the user.
Entity candidate and dialog state list help that process to find
the best calculated question template. Question templates are
also ranked, which helps us to restrict our choice. In case the
system couldn’t find the exact template match, it is capable to
advice the user which question template to use.
Now, when the appropriate question template is found, we
can generate an answer related either to an answer or a sub-
question template.
If entity in the template is represented as an external
information resource, we have to a find semantic description of
that source in our ontology. External information resources
could be a web service, a DLL library or a stored procedure in
a database. A specifically written software wrapper then calls
an appropriate method, which returns the result that represents
the required entity values.
Before the answer is shown to the user, a special process
records the user activity and alters the dialog state table. If
dialog with the user is not finished yet, an answer is formulated
as a question. At that stage we have entered in the question
answer dialog.
POGOVORNI SISTEM
DIALOG
STATES
TABLE
ENTITY
CANDIDATE
DETECTION
TEMPLATE
MATCHING
ANSWER
GENERATION
BASED ON
TEMPLATE
EXTERNAL
INFORMATION
SOURCES
DOMAIN SPECIFIC
KNOWLEDGE
(ONTOLOGY)
WRAPPER
ONTOLOGY
INSTANCES
TEMPLATES
USER ACTIVITIES
TRACKING
Fig. 2 Question answering process
VIII. CONCLUSION
This article describes our ontology-driven question
answering system with semantic web services support. While
Advances in Communications, Computers, Systems, Circuits and Devices
ISBN: 978-960-474-250-9
201

we didn’t want to build large knowledge corpuses of Slovenian
language, we decided to semantically describe our domain-
specific knowledge. The key component to our system is a
well defined and semantically described ontology based
knowledge database. Although there are some methods for
storing ontologies, we built our own ontology mapping to
relational database.
Because the question answering system should somehow
understand natural language we managed to provide question
templates. Question templates are a bridge between sentences
and Ontology. The template matching process is the most
important part in our system. This process is responsible for
the entire conversation dialogue. The answer generation
process is also built on top of the question templates. Some
entities in those templates should be filled from ontology
instances or even from external knowledge resources like web
services. Our question answering system also tracks user’s
behavior.
A challenge for our future work is to improve algorithms for
entity candidate detection and to speed up the algorithm for
finding the minimum distance in question templates. A special
treatment will be given to expand the set of external resources.
Our ontology based knowledge database should always
grow. You can get good results only if you have a large
enough and quality knowledge corpus.
REFERENCES
[1] G. Antoniou, F. Harmelen, Web Ontology Language: OWL
[2] I. Čeh, M. Ojsteršek, Developing Question Answering System for the
Slovene Language, WSEAS transactions on information science and
applications, Issue 9, Volume 6, September 2009.
[3] The Protégé Ontology Editor and Knowledge Acquisition System,
http://protege.stanford.edu, visited on October 2010.
[4] N. Shadbolt, W. Hall and T. Berners-Lee, The Semantic Web Revisited,
IEEE Intelligent Systems, Volume 21, No. 3, 2006.
[5] E. Sneiders, Automated Question Answering Using Question Templates
That Cover the Conceptual Model of the Database, Proceedings of the
6th International Conference on Applications of Natural Language to
Information Systems-Revised Papers, 2002.
Advances in Communications, Computers, Systems, Circuits and Devices
ISBN: 978-960-474-250-9
202