Applied Mathematical Sciences, Vol. 9, 2015, no. 130, 6491 - 6505
HIKARI Ltd, www.m-hikari.com
http://dx.doi.org/10.12988/ams.2015.54361
Semantics Question Analysis Model for
Question Answering System
Syarilla I. Ahmad Saany1*, Ali Mamat2, Aida Mustapha2,
Lilly S. Affendey2 and M. Nordin A. Rahman1
1 Faculty of Informatics and Computing, Universiti Sultan Zainal Abidin
(UniSZA), Tembila Campus, Besut 22200, Terengganu, Malaysia
Corresponding author*
2Faculty of Computer Science and Information Technology, Universiti Putra
Malaysia (UPM), 43400 UPM Serdang, Selangor Malaysia
Copyright © 2015 Syarilla I. Ahmad Saany et al. This article is distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
Abstract
Question Answering (QA) system takes Natural Language (NL) questions and
output the exact answer extracted from Knowledge Base (KB). Handling NL
question requires semantic analysis and interpretation to obtain the potential
relevant answers from the KB. Providing users with the most relevant answers to
their questions is an issue. Many answers returned are not relevant to the
questions. The purpose of this paper is to present a semantic question analysis
model for question answering system that correctly interpret the user’s question in
order to yield correct answers. The Semantic Question Analysis Model exploits
the approach of User Modelling and Relevance Feedback. To measure the ability
of Semantic Question Analysis Model in analysing and translating user NL
question, the model is implemented in an ontology-based question answering
system (known as QAUF). Comparison of the results is done to determine the
success of the proposed model with the existing QA systems, AquaLog and
FREyA. The Semantic Question Analysis Model in QAUF shows a relatively
increased in the QAUF F-measure. The finding of this study demonstrates that
QAUF has a good precision percentage in returning relevant answers for each NL
question.
Keywords: Semantic Question Analysis Model Question Answering System
Ontology
6492 Syarilla I. Ahmad Saany et al.
1. Introduction
Question answering (QA) resides within Information Retrieval (IR) and Natural
Language Processing (NLP) research areas. Information retrieval (IR) was coined
in [3] as the process or method to pull out a list of document citation containing
the needed information from the storage. At this present, many changes and
improvements are done to IR which resulted the retrieval list may in images,
multimedia, web pages and accurate answers as opposed as a list of documents’
citation. Examples of IR systems are Web’s most popular search engines such as
Google, Bing or Yahoo. Any IR system aims to seek texts, images videos or any
types of medium that matched with user’s query [14]. The user’s query or also can
be known as user’s specified needs, is conveyed in keywords manner. However,
this user’s specified needs may encounter loss of semantic information. The IR
system only yields the most related (matched) links of the documents list. This
method could not retrieve the exact answer of the user’s query except the user has
to search, locate and extract the necessary and adequate information from the
returned documents as opposed to QA.
QA is an IR that involves NLP mechanisms. NLP was formed firstly with the aim
of studying problems in the automatic generation and understanding of natural
language in artificial intelligence and linguistic [2]. NLP creates an easy and
friendly interface known as natural language interface (NLI). NLI eliminates the
need to understand a particular query language or query syntax. Research in NLI,
such as in [1] [8] [20] [24], are exponentially fostered with the availability of
many NLP tools for text processing.
QA system is a system that takes NL questions and the output is either the small
text fragments containing the answer or the exact answer extracted from
texts/documents or Knowledge Base (KB) [17]. It captures the piece of
information from the repository (KB or corpus) that actually matches the user’s
information need. The goal is to return an accurate answer for any user’s NL
question. Essentially, QA system needs to have the ability to linguistically
analyse, interpret and process a NL question before an accurate answer can be
returned. This is not an easy task to automatically capture the semantics lies in a
complex question structure. In the context of this research a question is
considered to be simple if the answer is a piece of information that has been
located and retrieved directly as it appears in the information source. On the other
hand, a question is considered complex if its answer needs more on elaboration
[7]. The semantic information contained in the user Natural Language (NL)
question may miss or lose during the question analysis process.
Handling NL question requires semantic analysis, interpretation and
transformation into an executable query so that the potential answers can be
obtained from the corpus or knowledge base. The purpose of this paper is to
present a semantic question analysis model for question answering system.
Semantics question analysis model 6493
The proposed of Question Analysis Model (QAM) system model employs the
approach of UM and RF approach. The remainder of this paper is prepared as
follows: Section 2 reviews the related works in QA system. The proposed model
and result analysis are presented in Section 3. Section 4 sums up the paper with
conclusions and future works.
2. Related Works
Generally, the QA systems consist of three main modules: question analysis,
document/information processing and answer. User’s NL question is analysed and
processed. The QA system has to understand exactly what the question is in order
to extract the answer from the text corpus or knowledge base (KB). The keyword
(focus) and the answer type of the question is taken from the extracted NL
question. The keyword (focus) of the question is used as the input to other
modules in QA system and an answer type is a specification of type of entity that
would comprise a potential answer to the question. Figure 1 illustrates the general
components of QA system.
Figure 1: General Components for QA system
A question-answering (QA) system takes in user questions in natural language
(NL). This NL question is handled through three stages, which are question
analysis, document processing, and answer processing. Other supporting
components to a QA system include the knowledge base (KB), ontology and
WordNet. The document analysis component obtains some potential answer and
throws to the answer processing component.
noitseuQ
gnesoiisuP
esnmoutQ
gnessoisuP
Text
Fragments
/ Passage Sources of
Answer
Answer
esnmouti
BK
NL Question
uiwonQ
gnesoiisuP
Query
6494 Syarilla I. Ahmad Saany et al.
Traditionally, QA systems run on text document collections to get answers that
match with user’s query. In recent QA systems which are ontology-based QA
systems use Knowledge Bases (KBs) as their sources of answers [6] [12] [21].
This can be seen in the works of [5] [13] [14] [16] [21] and many more. During
the early time, many QA systems used ontology as a mechanism to support query
expansion [21]. In recent studies of QA systems, KB is exploited not only for
answer searching but also used in mapping and transforming user’s NL question
into query representation.
In general, ontology provides general understanding of the domain knowledge,
common vocabulary used and specific definition of the relation between these
vocabularies. Knowledge of the domain is described by using concepts-based
[10]. In recent, improvements in semantic capability of QA have become one of
the research focuses. Many research have shown that ontology promotes the
semantic capability of a QA system which may eliminate the need to build
heuristics to recognize name entities and question classification [5] [14] [21].
AquaLog is the first ontology-based QA system that relies on the knowledge
encoded in the underlying ontology and its explicit semantics for question
analysis and answer retrieval [20]. In AquaLog, there are two components which
are the Linguistic Component (LC) and the Relation Similarity Service (RSS)
component. LC transforms the NL-question into NL query-triple format form.
The RSS component maps the query-triple format with ontology elements
generating the Ontology-Compliant queries known as Onto-Triples. AquaLog
only obtained 48.68% correct answers out of 76 total questions. This precision
score has opened up many research issues. One of the drawbacks is, AquaLog
failed to analyse and process user’s question containing modifier term. This
contributes to the lower percentage in the precision score obtained.
The stochastic syntax-parse model named Lexical Semantic Frame (LSF) is
adopted in [14]. The basis of this model is on analysing semantic probability
through the function of semantic relation, P (LSF | wi … wj) where wi is among
words of character string s. Both words in corpus and the Hyponym part-of-
speech (POS) information are used in estimating semantic probability, P (LSF | wi
… wj). They apply the pattern matching technique on user’s question with the
ontology. They also introduce a novel question classification hierarchy that is
based on answer type and question pattern. During the situation of no answer, the
system will expand the ontology in an attempt to provide the answer.
Nevertheless, the consistency and possibility of self-produced knowledge based
on ontology semantic relations have become one of the drawbacks.
FREyA is a Feedback Refinement and Extended Vocabulary Aggregation QA
system which combines syntactic parsing with knowledge in ontology to reduce
the customization effort [6]. To understand the user’s question, this system relies
on knowledge encoded in the ontology. The user’s NL question is translated into
Semantics question analysis model 6495
the set of Ontology Concepts (OCs). The user question is mapped to the Ontology
Concept (OC) automatically or semi-automatically. However, different users may
interact differently with the suggestion given by FREyA for the ambiguous
question phrases or terms. This makes FREyA’s method requires heavy
supervision to be utilized. FREyA always returns the answer to user’s question
although the answers returned are partial correct or incorrect. FREyA obtains
92.4% of both recall and precision values [5].
The intention of this research is to semantically analyse, interpret and transform
user question into executable queries for a better performance QA system. Here, a
better performance QA means the system is able to return a correct answer based
on user’s intent question. The first essential task is the understanding of the users’
needs and users’ information seeking behaviour. The second essential task is the
analysing and processing of the users’ needs expressed in a question (request).
Lastly, the third task is providing a strategy for matching of the user’s question to
data or information on the document collections or knowledge base.
The investigations of this paper are embedded in the following research questions:
i. How users’ needs and information seeking shall be understood from the user’s
NL question?
ii. How shall users’ needs that are expressed in a question be analysed, interpreted
and processed?
iii. How shall the user’s question be matched for the answers on the knowledge
base?
3. The Semantic Question Analysis Model
The general context of a question-answering (QA) system formalizes U, as user’s
natural language (NL) query. U holds a specific condition of retrieving a non-
empty set of answer from a knowledge base; here, a knowledge base comprises a
set of information labeled as I = {i1, i2, …, in} and a set of Si I, consists of
information that is totally (or partially) relevant to the user’s requirement [18]19].
To yield Si, the QA system depends on the user requirements and all information
in the knowledge base, which denotes in the form of function f: (U, I).
In this paper, Semantic Question Analysis Model is implemented in an ontology-
based QA system called QAUF (Question Answering system with User Modeling
and Relevance Feedback), which models the user requirement, U, as a
combination of three elements: NL query, user modeling, and relevance feedback.
This is denoted as U = (Unl-query, Uuser-modeling, Urelevance-feedback). Within this context,
[18] [19] defined each element as:
Unl-query is the goal/answer to be retrieved
Uuser-modeling is a set of user interest, defined as Uuser-modeling = {knowledge
base concept, question context, language theory}
6496 Syarilla I. Ahmad Saany et al.
Urelevance-feedback is a set of interactions with the system Urelevance-feedback =
{modification of term weight, query expansion, query simplification}
A complete function of the proposed QA system model is defined in [18] [19] as
follows:
f: (Unl-query, Uuser-modeling, Urelevance-feedback, I) Si (1)
In the following sub-section, details explanation of the question analysis process
will be presented.
3.1 The Semantic Question Analysis Model and Process Flow
Generally the combination of User Modelling (UM) and Relevance Feedback
(RF) in question analysis model will allow additional information to be obtained
from the user and user’s NL question [18][19]. Additional information is needed
to enrich the information description of the question and to avoid any ambiguities
in user’s question. As a result, this contributes to the correct and accurate answer
yielded. Figure 2 depicts that the Semantic Question Analysis Model.
Figure 2: The Semantic Question Analysis Model [18][19].
The above model uses user modelling to acquire user interest and profile in
categorizing the answers. Relevance feedback is utilized in the second phase of
question analysis process in order to better up the user’s question specifications.
User’s NL question consists of words which also referred as terms in this
research. Terms of user’s question are extracted in order to proceed with further processes of finding the correct answer. Based on Figure 2, terms focus, head focus,
head focus, modifier of
head focus, focus complement, user
profile
User modeling process
<obtaining user interest and
profile greedily>
Relevance feedback process
<query modifying
mechanism>
Knowledge
base
Ontology
WordNet
Lookup catalog
<New query vector>
Semantics question analysis model 6497
modifier term, modified term, focus complement and user profile become input to
the Question Analysis Model. This model considers four (4) lexical elements
which are focus, head focus, modifier term and modified term. Example and
details explanation of lexical elements terms are illustrated in [18] [19].
The user’s NL question is also extracted its triple which is in the form of <subject,
verb, object>. A number of pre-requirements components are also needed such as
knowledge base, ontology, WordNet and Modifier Lookup [18] [19].
3.2 Identifying and Extracting Terms
In this model, a user’s NL question is received and analysed. The submitted NL
question is denoted using vector space model as follows:
Q = {(T1, W1), (T2, W2), …, (Tn, Wn)} (2)
Where, T represents the terms that exist in the question which include
{head_focus, modifier_head_focus, focus_complement}. W is the weight of T.
The weight of each term is calculated using the formula tf × idf where tf is the
term frequency in knowledge base and idf is the inverse term frequency in the
whole knowledge base collections.
First, the NL question will be tokenized and parsed for its Part-Of-Speech (POS)
using Tokenizer and POS tagger respectively. NL question such as “Could you
tell me what is the highest point in the state of Oregon?” will be tokenized and
parsed. In this research, NL question is based on the Raymond Mooney gold
standard dataset. The selected NL questions that are used for the experiment must
contain the modifier term. Language rules are exploited in identifying and
extracting the substantial terms of NL question. Later, an initial user profile will
be auto-generated using a language model.
3.3 Applying User Modelling (UM) in User’s Question
UM is applied during question analysis in order to enrich user’s question with
additional terms. This is to ensure that all the terms that carry semantic meanings
of the original user’s question are regarded. In constructing the user profile, a set
of knowledge sources needs to be identified. There are three (3) methods used in
constructing user profile attributes: default assumptions, assumption and dialog
contribution from system [23]. Based on these methods, the model intends to
yield user interest based on three (3) aspects using assumption methods:
knowledge base concept, question context and language theory.
As mentioned previously, an initial user profile will be auto-generated based on
question context using language model. NL user’s question and the initial user
profile will be used to produce new query. In UM process, the modifier terms will
6498 Syarilla I. Ahmad Saany et al.
be searched for their semantic similarity using WordNet, Knowledge Base,
ontology and Modifier LookUp. These terms will be added to the initial profile
which will be used to generate new query. The new query is represented in VSM
as follows:
Q’ = {(C1, W1), (C2, W2), …(Cn, Wn)} (3)
Where, C denotes the instances to the KB class concept of query or terms and W
is the weight of C. Figure 3 summarizes the steps of the UM phase.
Algorithm : User Modelling phase
Input : user profile, knowledge base, assumption from question context
and language theory, question (Q)
Output : A query in vector space form (Q’)
1. Start
2. Q’
3. If (user profile is available) then
i. Construct the possible Q’ that related to Q,
Q’ Q’ + (ci, wi)
4. Initialize knowledge base for ontology analysis
i. Generate Q’, Q’ Q’ + (ci, wi)
5. Apply the assumption method based on query context and language theory
i. Generate Q’, Q’ Q’ + (ci, wi)
6. End
Figure 3: Algorithm of Applying UM on User’s Question
As an example, the user’s NL question: “What are the cities of the state with the
highest point?” gives Unl-question = {city, state, highest, point}.The items of Unl-
question are terms of the lexical elements. User question also has the triple format as
<city, are, state>. This triple is mapped to the concepts and relationships from the
knowledge base. As for the subject, city is mapped to any corresponding class. For
the verb, are is mapped to any corresponding property in the knowledge base
which has relationship with the class city.
3.4 Applying Relevance Feedback (RF) in User’s Query
In the RF process (Phase 2), the model considers three (3) aspects of modifying
the query representation (query generated from phase 1). They are modification of
term weight, question expansion and question simplifying. In Phase 2, based on
user’s question, if necessary, user is allowed to modify the weight associated to
Semantics question analysis model 6499
the question term. This will contribute to the formation of the new query vector.
To avoid ambiguities in user’s question, additional information is also acquired in
regards of the question entered through inserting new terms. New terms are
suggested by consulting WordNet, Knowledge Base, Ontology and Modifier
LookUp. At the end of Phase 2, another new following set of solutions (new query
vector) will be produced which represented in VSM:
𝑄" = {𝑞1" , 𝑞2
" , 𝑞3" … , 𝑞𝑛
" } (4)
Thus, the set of solutions can be represented as follows:
qi”= {{(T1,i, ∆W1,i), (T2,i, ∆W2,i), …, (Tn,i, ∆Wn,i)},{ (Tn+1,i, Wn+1,i),
(Tn+2,i, Wn+2,i), …, (Tk,i, Wk,i)}} (5)
Where, ∆W represents the change of weight value, and Tn+1,i, Tn+2,i, Tk,i are the
new inserted terms that associated with their weight Wn+1,i, Wn+2,i, Wk,I
respectively and i = 1, 2, 3,…n and k > n [19].
To determine the best solution to be used, the initial NL question Q is compared
with all solutions 𝑄′′ and the similarity scores are computed. The similarity score
formula used is as follows:
)
1()",( imW
mWQQSim
k
m
(6)
The best possible of 𝑞𝑖"(that is the nearest similar to Q) is selected from all
possible Q’’. Here,𝑞𝑖"that has the highest similarity score is recognized as nearest
similar to Q. If 𝑄𝑛𝑒𝑤 is the new query vector that is submitted for retrieving
answer, thus𝑄𝑛𝑒𝑤 can be denoted as follows:
𝑄𝑛𝑒𝑤 = (𝑟𝑢𝑙𝑒, ℎ𝑖𝑔ℎ𝑒𝑠𝑡(𝑞𝑖")) (7)
where, 𝑟𝑢𝑙𝑒 is the rule obtained from the Modifier Lookup during the stage of
applying UM technique and will be applied to the query containing the modifier
and modified terms and ℎ𝑖𝑔ℎ𝑒𝑠𝑡(𝑞𝑖”) is the nearest similar query vector that
generated in the stage of applying RF technique. Figure 4 summarizes the steps of
the RF phase
6500 Syarilla I. Ahmad Saany et al.
Algorithm : Relevance Feedback phase
Input : knowledge base, WordNet, ontology, Modifier Lookup, question (Q)
Output : A set of modified query in vector space form (Q”)
1. Start
2. Flag1 1
3. While (Flag1) // generating of i possible Q”
i. i 1
ii. 𝑄" ← 𝑄
iii. Map with knowledge base for terms weight modification
a. 𝑄" ← ∆𝑄
iv. Flag2 1
v. While (Flag2) // inserting of k possible new terms
a. k 1
b. Map with WordNet, ontology and Modifier Lookup for
new terms
insertion - 𝑞𝑖" ← 𝑞𝑖
" + (𝑇𝑘, 𝑊𝑘)
c. k k + 1
d. If (no more possible new terms can be inserted)
then - Flag2 0
vi. ii + 1
vii. If (no more possible Q” can be generated) then
a. Flag1 0
4. End
Figure 4: Algorithm of Applying RF Technique
4. Experimental Result
This research uses the Raymond Mooney dataset in Geobase ontology of United
States geographical information which has a total of 880 annotated user questions
dataset [11]. A review of relevant literature has shown that the Raymond Mooney
dataset is extensively used for the evaluation of ontology-based QA systems and
other natural language interfaces systems [4][5][16][22]. Out of 880 questions
dataset, 607 questions containing modifier terms will be used in this study.
To measure the ability of Semantic Question Analysis Model in analysing and
translating user NL question, the model is implemented in an ontology-based
question answering system (known as QAUF). Comparison of the results is done
to determine the effectiveness of the proposed model as compared to the existing
QA systems, AquaLog [21] and FREyA [6]. This is to determine the ability of
Semantic Question Analysis Model in analysing and interpreting user’s NL
question into executable query which resulting an accurate returned answers. The
Semantics question analysis model 6501
evaluation metrics are similar to those used in [6][21], which are the percentage of
quantitative retrieval performance recall and precision [15].
The experimental settings for this research are as follows:
i. Experiment A:
Experiment A is to test the performance of AquaLog [21] using the gold standard
dataset.
ii. Experiment B:
Experiment B is to test the performance of FREyA [6] using the gold standard
dataset.
ii. Experiment C:
Experiment C is to test the effectiveness of user model and user feedback in
question analysis process. The objective of this experiment is to quantify the
accuracy of returned answer by incorporating user model and user’s relevance
feedback during question analysis process. User model and user’s relevance
feedback are the experimental parameters which will be evaluated by determining
the accuracy of returned answer when additional information is provided to
disambiguate any ambiguities during question analysis process.
These experiments are using the same group of dataset i.e. the Raymond Mooney
question dataset over US geography ontology and knowledge base (KB). Table 1
summarizes the overall experiment results on 607 questions from Raymond
Mooney Geoquery dataset. The experimental results show significantly impacts
on the success of 90 % relevant and correct returned answers as shown in Table 1.
Table 1: Overall Performance of QAUF Average
R
AP
F-Measure MAP
Applying UM
and RF 0.947 0.947 0.947 0.9
This question analysis model generated 94.7 % average recall and average
precision. This is due to several users’ questions that cannot be analysed and
processed by the proposed model such as 'Which capitals are not major cities?',
'What is the lowest point in Nebraska in meters?', 'How many square kilometres in
the US?', 'What is the average population per square km in Pennsylvania?' and
'What is the longest river that passes the states that border the state that borders
the most states?'.
Experimental evidence shows that question that falls under negation-type question
(e.g.: 'Which capitals are not major cities?'), question contains explicit
measurement unit (e.g.: 'What is the average population per square km in
Pennsylvania?' ) and complex question that requires more than two level in
comparative or evaluative processing (e.g.: 'What is the longest river that passes
the states that border the state that borders the most states?') failed to get relevant
and correct answer from QAUF.
6502 Syarilla I. Ahmad Saany et al.
Table 2: Performance Comparison on QAUF, AquaLog and FREyA
Average R AP F-
measure
QAUF 0.947 0.947 0.947
AquaLog 0.43 0.42 0.42
FREyA 0.924 0.924 0.924
From Table 2, AquaLog shows a relatively poor performance in comparison of
FREyA and QAUF. This is because, AquaLog does not have the facility to
process the type of how much/how many questions and questions contain
modifier. Whereas, FREyA, generated slightly lower in precision as compared to
QAUF. Experimental evidence suggests that the proposed question analysis model
outperforms AquaLog and FREyA.
In the Semantic Question Analysis Model, UM is utilized for filtering and
classifying the answers of user NL question based on the KB concept, question
context or/and the language theory. Then, RF is employed in modifying the query
representation based on the term’s weight, query expansion or query simplifying.
5. Conclusion and Future Works
The extracted lexical of the user’s question may contain some terms that influence
the correctness of the answer returned. This term is referred as modifier term. The
main objective of this research is proposing a new a question analysis model to
correctly interpret all the modifier terms of the user’s question in order to yield
correct answers in the prototype of QAUF (Question Answering system with User
modelling and relevance Feedback).The Semantic Question Analysis Model in
QAUF shows a relatively increased in the F-measure where QAUF is 94.7%,
FREyA is 92.4% and Aqualog is only 42%. The finding of this study
demonstrates that QAUF has a good precision percentage in returning relevant
answers for each NL question.
In the near future, QAUF aims to implement an automatic modification,
expansion and simplifying of the user’s NL question by exploiting the machine
learning mechanism.
Acknowledgements. Special thanks to Universiti Sultan Zainal Abidin (UniSZA),
Centre for Research and Innovation Management (CRIM), UniSZA, Universiti
Putra Malaysia (UPM) and Management of Faculty of Informatics and Computing
(FiK), UniSZA for their supports in this research.
References
[1] A.M. Popescu, O. Etzioni, & H. Kautz, Towards A Theory of Natural
Language Interfaces to Databases, International Conference on Intelligent User
Interfaces, (2003), 149-157. http://dx.doi.org/10.1145/604045.604070
Semantics question analysis model 6503
[2] A. Turing, Computing Machinery and Intelligence, Mind, LIX (1950), no.
236, 433 – 460. http://dx.doi.org/10.1093/mind/lix.236.433
[3] C.N. Mooers, Zatocoding Applied to Mechanical Organization of Knowledge,
American Documentation, 2 (1951), no. 1, 20-32.
http://dx.doi.org/10.1002/asi.5090020107
[4] C. Wang, M. Xiong, Q. Zhou, & Y. Yu, PANTO: A Portable Natural
Language Interface to Ontologies, European Semantics Web Conference
(ESWC'07), 4519 (2007), 473-487.
http://dx.doi.org/10.1007/978-3-540-72667-8_34
[5] D. Damljanovic, M. Agatonovic, & H. Cunningham, Natural language
interface to ontologies: Combining syntactic analysis and ontology-based lookup
through the user interaction, Proceedings ESWC-2010, Part I, LNCS, 6088 (2010),
106-120. http://dx.doi.org/10.1007/978-3-642-13486-9_8
[6] D. Damljanovic, M. Agatonovic, and H. Cunningham, FREyA: an Interactive
Way of Querying Linked Data Using Natural Language, Proceedings of the
European Semantic Web Conference, (2012), 125-138.
http://dx.doi.org/10.1007/978-3-642-25953-1_11
[7] D. Mollá-Aliod, & J. Vicedo, Question answering, Indurkhya and Damerau
(eds) Handbook of Natural Language Processing, (2010), 485-510.
[8] D. Moldovan, S. Harabagiu, R. Girju, P. Morarescu, F. Lacatusu, A. Novischi,
et al., LCC Tools for Question Answering, Proceedings of TREC, (2002).
[9] E.M. Voorhees, The TREC question answering track, Natural Language
Engineering, 7 (2001), no. 4, 361-378.
http://dx.doi.org/10.1017/s1351324901002789
[10] J. Heflin, OWL Web Ontology Language Use Cases and Requirements,
(2004). http://www.w3.org/TR/2004/REC-webont-req-20040210/
[11] MooneyData, Retrieved from.
http://www.cs.utexas.edu/~ml/nldata/geoquery.html
[12] O. Fernandez, R. Izquierdo, S. Ferrandez, and J.L. Vicedo, Addressing
Ontology-based question answering with collections of user queries, Journal of
Information Processing and Management, 45 (2009), no. 2, 175-188.
http://dx.doi.org/10.1016/j.ipm.2008.09.001
[13] P. Atzeni, R. Basili, D. H. Hansen, P. Missie, P. Paggio, M.T. Pazienza, et
al., Ontology-based Question Answering in a Federation of University Sites: The
6504 Syarilla I. Ahmad Saany et al.
MOSES Case Study, 9th International Conference on Applications of Natural
Language to Information Systems (NLDB '04), Manchester (United Kingdom),
Lecture Notes in Computer Science, 3136 (2004), 413-420.
http://dx.doi.org/10.1007/978-3-540-27779-8_40
[14] Q. Guo, & M. Zhang, Question Answering System Based on Ontology and
Semantic Web, Third International Conference of Rough Set and Knowledge
Technology 2008, Chengdu, China, 5009 (2008), 652-659.
http://dx.doi.org/10.1007/978-3-540-79721-0_87
[15] R. Baeza-Yates, & B. Ribeiro, Modern Information Retrieval, Harlow:
Addison-Wesley (1999).
[16] R. Iqbal, M.A.A. Murad, M.H. Selamat, & A. Azman, Negation Query
Handling Engine For Natural Language Interfaces to Ontologies, International
Conference on Information Retrieval and Knowledge Management (CAMP '12),
(2012), 249-253. http://dx.doi.org/10.1109/infrkm.2012.6204983
[17] S. Harabagiu, D. Moldovan, C. Clark, M. Bowden, A. Hickl, P. Wang,
Employing Two Question Answering Systems in TREC-2005, Proceedings of the
Fourteenth Text REtrieval Conference, (2005).
[18] S. I. A Saany, A. Mamat, A. Mustapha, L. S. Affendey, A Strategy for
Question Interpretation in Question Answering System, International Journal of
Computer Science and Telecommunications, 4 (2013), no. 5, 38-43.
[19] S.I.A. Saany, A. Mamat, A. Mustapha, L.S. Affendey, Incorporating User
Modeling and Relevance Feedback in Question Analysis Model, Proc. in 2nd 2013
IEEE Conference on Control, Systems and Industrial Informatics (ICCSII),
Bandung, Indonesia, (2013), 113- 118.
[20] T. Strzalkowski, J.P. Carballo, J. Karlgren, A. Hulth, P. Tapanainen, & T.
Lahtinen, Natural Language Information Retrieval: TREC-8 Report, In TREC,
(1999).
[21] V. Lopez, V. Uren, E. Motta, & M. Pasin. AquaLog: An Ontology-driven
Question Answering System for Organizational Semantic Intranets, Journal of
Web Semantics, 5 (2007), no. 2, 72-105.
http://dx.doi.org/10.1016/j.websem.2007.03.003
[22] V. Tablan, D. Damljanovic, & K. Bontcheva, A Natural Language Query
Interface to Structured Information, European Semantic Web Conference (ESWC
2008), 5021 (2008), 361-375.
http://dx.doi.org/10.1007/978-3-540-68234-9_28
Semantics question analysis model 6505
[23] W. Wahlster, & A. Kobsa, User Models in Dialog Systems, Heidelberg:
Springer, (1989). http://dx.doi.org/10.1007/978-3-642-83230-7
[24] X. Li, & D. Roth, Learning Question Classifiers, Proceedings of the 19th
International Conference on Computational Linguistics - Volume 1 (COLING
'02), Association for Computational Linguistics, Stroudsburg, PA, USA, 1 (2002),
1-7. http://dx.doi.org/10.3115/1072228.1072378
Received: April 15, 2015; Published: November 2, 2015