ontology-based question answering system

8/3/2019 Ontology-Based Question Answering System

1/3

Ontology-based Question Answering System

Sheen-Mok Lee, Pummo Ryu, Key-Sun Choi

Korea Advanced Institute of Science and Technology, 335 Gwahangno

(373-1 Guseong-dong), Yuseong-gu, Daejeon 305-701, Republic of Korea

HTU

{smlee, pmryu}@world.kaist.ac.krUTH

,HTU

[email protected]

Abstract. In this paper, we construct IT-domain question-answering system as

an application system of the IT-domain ontology. We defined 16 types of

questions based on question type definition by DARPA. For each of thequestion type, an inferencing method is designed and implemented. We

demonstrate the user interface which can connect users with the QA system and

the ontology.

Keywords: question-answering(QA), ontology, question type

1 Introduction

Question answering has been vividly researched since 1999 when TREC(Text

Retrieval Conference) started to construct QA test collection and evaluate systems. As

a new approach for QA, ontology-based QA system has following advantages otherthan improved performance. i) offering additional information about an answer, ii)

providing measures of reliability, iii) explaining how the answer was derived[3].In order to deal with various kinds of inferencing scheme, we used a taxonomy of

16 question types, which was defined by Graesser and Peterson(1994) and introduced

by DARPA QA white paper[1]. We implemented ontology lookup method for each of

the 16 question type. We also constructed user interface that will make it easy for

users to interact with QA system and show how our ontology works for the question.

2 QA Processing Unit: Triplet

In our approach, the unit for I/O and internal processing is not a NL sentence but atriplet, which can be directly applied to ontology lookup, shown in table 1.

Table1. Types and examples of triplet for question answering

Type lexical triplet ontological triplet

question

-type

(is used, time division

switch, why)

GOAL_ORIENTATION(isUsedFor,

time_division_switch, ?X)

answer-

type

(is used, time division

switch, multiplexing)

GOAL_ORIENTATION(isUsedFor,

time_division_switch, multiplexing)


2/3

In table1, we showed triplets which are necessary to answer Why is time division

switch used?. After question-type lexical triplet is input to the system, it is converted

to question-type ontological triplet. Through ontology lookup, we get answer-type

ontological triplet, which is converted to answer-type lexical triplet output to user.

3 Architecture

PropertySearch

InputQuery Triplet

InstanceSearch

Instance/ClassDiscrimination

Concept Completion

Concept Completion/Instance

Property ValueSearch AnswerSelection

AnswerGeneration

AffiliatingClass Search

Class Search

Concept Completion/Class

RepresentativeInstance Search

AnswerVerification

Yes

No

ExampleSubclassSearch

Enablement/Goal Orientation

Target ClassSearch

Dont knowMessage

QA end

Figure1. Architecture of QA system

In figure1, we show the path for ontology lookup for 4 types(concept completion,

example, enablement, and goal orientation). For each of the type, we can see diverse

search paths. For example, after searching a class for questioned entity, three kinds ofpaths are possible. The path selection is determined by query type.

In determining search path, we should also consider the type of questioned

entity(class/instance). In figure1, we can see that the path for concept completion type

is divided with two paths according to the result of class/instance discrimination. If

user is asking about an instance like Samsung Electronics, we should search the

instance in the next step, while if he is asking about a class like DMB service, we

should search the class in the next step.

References

1. J. Burger et al : Issues, Tasks and Program Structures to Roadmap Research in Question &

Answering (Q&A). NIST. 2001.

2. D. Kelly and J. Lin: Overview of the TREC 2006 ciQA Task. In TREC 2006.

3. D. Mc Guinness: Question Answering on the Semantic Web. IEEE Intelligent Systems,

19(1), 2004.

4. J. Allen : Natural Language Understanding. The Benjamin/Cummings Publishing Company,

Inc., 1995, pp 414-419.

5. V. Lopez and E. Motta: Ontology-driven Question Answering in AquaLog. In Proceedings

of 9PthP international conference on applications of natural language to information systems,

2004.


3/3

Demo Script for Ontology-based QA System

1. Functionalities of User Interface for Ontology-based QA System

A. Our user interface consists of following two windows.

B. Question Answering window

i. Providing users with the queries that users can get the response

from ontology-based QA system

ii. Supporting the user process of question and answering

C. Ontology Search Path windowi. Graphically visualizing the search path of a query

2. Demo Sequence

A. Select a question type of the query to ask (selecting WHAT).

B. Select a lexical form relation of the query to ask (selecting jiwi(position)).

C. By clicking Search Query List button, get the possible query list.

D. Select a query to ask (selecting [PyeongMoo Park, jiwi(position),

WHAT] which means What is the position of PyeongMoo Park?).

E. By clicking Get Answer button, get the answer and graphically

observe the search path in the ontology.(Figure2)

F. There are many additional functionalities that users can get additional

information of ontology in the ontology search path window.

Figure2. User interface after receiving answers from QA system

QA modelsQA typesRelations

Query List

Query Searching Button

Answer Generation Button

System Answer

What is the position of Mr.

Park?

Hes a vice president of association

Hes a managing director.

Position Relations

Classes

Instances

QA window

Ontology search pathwindow

Names

ontology-based question answering system

Documents