Biomedical Question Types Classification using Syntactic and Rule

based Approach

Mourad Sarrouti

, Abdelmonaime Lachkar

and Said El Alaoui Ouatik

Department of Computer Science, FSDM, Sidi Mohamed Ben Abdellah University, Fez, Morocco

Department of Electrical and Computer Engineering, ENSA, Sidi Mohamed Ben Abdellah University, Fez, Morocco

Keywords: Biomedical Question Answering System, Biomedical Question Types Classification, Answer Types

Classification, Natural Language Processing, Syntactic Patterns, Unified Medical Language System.

Abstract: Biomedical Question Types (QTs) Classification is an important component of Biomedical Question

Answering Systems and it attracted a notable amount of research since the past decade. Biomedical QTs

Classification is the task for determining the QTs to a given Biomedical Question. It classifies Biomedical

Questions into several Questions Types. Moreover, the Question Types aim to determine the appropriate

Answer Extraction Algorithms. In this paper, we have proposed an effective and efficient method for

Biomedical QTs Classification. We have classified the Biomedical Questions into three broad categories.

We have also defined the Syntactic Patterns for particular category of Biomedical Questions. Therefore,

using these Biomedical Question Patterns, we have proposed an algorithm for classifying the question into

particular category. The proposed method was evaluated on the Benchmark datasets of Biomedical

Questions. The experimental results show that the proposed method can be used to effectively classify

Biomedical Questions with higher accuracy.

1 INTRODUCTION

Unlike Information Retrieval (IR) System that

retrieve a large number of documents that are

potentially relevant for the questions posed by the

inquirers, Question Answering (QA) System aims to

provide inquirers with direct, precise answers to

their questions, by employing Information

Extraction (IE) and Natural Language Processing

(NLP) methods (Athenikos and Han, 2009).

Typically an automated QA System consists of

three main phases (Athenikos and Han, 2009;Gupta

and Gupta, 2012): Questions Processing, Documents

Processing and Answers Processing phases. Figure 1

illustrates the generic architecture of Biomedical QA

System.

Biomedical QTs Classification is a crucial task

of any Biomedical QA System. It classifies

Biomedical Questions into several Biomedical QTs.

In addition, the main goal of Biomedical QTs

Classification is to determine the Excepted Answer

Type to a given Biomedical Question, such as

whether the answer should be a Biomedical Entity

Names, short text summarization, paragraph or just

“Yes” or “No”.

Documents Processing

Questions

Questions Processing

Query Reformulation

Question Classification

Documents

And Passa

Candidate

Passages

Answers Processing

Answer

(

)

Documents

Figure 1: Generic Architecture of Question Answering

System.

Indeed, in order to extract the answer for a given

Biomedical Question, the system should know in

advance the Excepted Answer Type that allows a

Biomedical QA System to use type-specific answer

retrieval algorithms and to reject possible answers of

the wrong type. Therefore, Biomedical QTs

Classification task plays a vital role in Biomedical

QA System that can strongly affect positively or

negatively the Answers Processing phase and hence

Sarrouti, M., Lachkar, A. and Ouatik, S..

Biomedical Question Types Classiﬁcation using Syntactic and Rule based Approach.

In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 1: KDIR, pages 265-272

ISBN: 978-989-758-158-8

265

determines the quality and overall performance of

the Biomedical QA System.

In recent years, several works have been done in

this field for open domain QA System (Zhang et al.,

2007; Tomuro, 2004), as well as for Biomedical QA

System. Ely et al. (1999,2000,2002) have proposed a

generic taxonomy of common Medical QTs and

another taxonomy which classifies questions into

Clinical vs Non-Clinical, General vs Specific,

Evidence vs No Evidence, and Intervention vs No

Intervention. For instance, question taxonomies have

some expressiveness limits. For example, Ely et al.,

(2000)’s question taxonomy provides only some

forms of expression for each question category,

when in the real world we may often retrieve several

other expressions for the same categories.

In fact, there are important factors that

distinguish Biomedical QA System from open-

domain QA System. Those factors include: (1) size

of data, (2) domain context, and (3) resources.

Indeed, in biomedical domain QA, the domain of

application provides a context for the QA process.

This involves domain-specific terminologies and

domain-specific types of questions, which also differ

between domain experts and non-expert users.

Athenikos and Han (2009) report the following

characteristics for restricted-domain QA in the

biomedical domain: (1) large-sized textual corpora,

(e.g., MEDLINE) , (2) highly complex domain

specific terminology, that is covered by domain-

specific lexical, terminological, and ontological

resources, (e.g., Unified Medical Language System

(UMLS)), (3) tools and methods for exploiting the

semantic information (e.g., MetaMap), and (4)

Domain-specific format and typology of questions.

Therefore from the above remarks, the

Biomedical QTs Classification block needs its own

methods that are different from others used for open

domain QA System that have been proposed for the

following Question Types: Location, Date, Person,

Organization, etc (Khoury, 2011).

In light of this, there are two main approaches of

Questions Classification: Syntactic Patterns-based

approach and Machine Learning. Due to the limited

number of Question Types and due to the lack of

quantity of labeled data, the Syntactic Patterns-based

methods have become the most popular methods in

QTs Classification System (Sung et al., 2008).

As far as we know

, there are no studies that have

discussed QTs Classification problem in the biomedical

domain and that is why several Biomedical QA System

that have been presented deal only with one Question

Type

(Weissenborn et al., 2013; Yang et al., 2015).In

this paper, we have proposed an efficient and

effective method for Biomedical QTs Classification.

Indeed, we have taken into account all Biomedical

QTs. We have defined the Syntactic Patterns of

each Biomedical QTs. In particular, we have

classified the Biomedical Questions into three broad

categories: Yes/No, Factoid and Summary Questions.

We denote that all the Excepted Answer Types are

summarized by these categories (Tsatsaronis et al.,

2012).

 Yes/No Questions: These are questions that

require “Yes” or “No” as an answer. For example,

“Is COL5A2 gene associated to ischemic heart

disease?” is a Yes/No Question.

 Factoid Questions: These are questions that

require a particular one or more of Biomedical

Entity Names (e.g., of a disease, drug, gene, a list of

gene names, etc.), a number(s), or a similar short

expression as an answer. For example, “Which genes

have been found mutated in Gray platelet syndrome

patients?” is a Factoid Question.

 Summary Questions: These are questions that

can only be answered by phrase extracted from

relevant document or by producing a short text

summarizing the most prominent relevant

information. For example, “What is the role of

anhedonia in coronary disease patients?” is a

Summary Question (Tsatsaronis et al., 2012).

The rest of this paper is organized as follows: In

Section 2, we review a representative sample of

work done on the task of QTs Classification. Our

proposed method for Biomedical QTs Classification

is presented in Section 3. The experimental results

are presented and discussed in Section 4. Finally, we

conclude the paper and describe future work in

Section 5.

2 RELATED WORK

Although research on QA System has boomed in

recent years, Question Classification has been a

large part in the research community of text mining

after the introduction of QA Track in the Text

REtrieval Conference (TREC) in 1999 as well as the

presentation of Biomedical QA in the BioASK

(Tsatsaronis et al., 2012). Several works that have

been presented for open domain QTs Classification

are usually based on Syntactic Patterns or rule-based

approach such as (Prager et al., 1999; Khoury, 2011;

Haris and Omar, 2012; Biswas et al., 2014).

However, laborious researcher’s effort is

required to create these rules. Some researchers have

used machine-learning approach. Li and Roth (2002)

have presented a hierarchical classifier for open

KDIR 2015 - 7th International Conference on Knowledge Discovery and Information Retrieval

266

domain QTs Classification based on the Sparse

Network of Winnows .Two classifiers were involved

in this work: the first one classified questions into

the coarse categories; and the other classified

questions into fine categories. Several syntactic and

semantic features were extracted and compared in

their experiments. Their result showed that the

hierarchical classifier achieved an accuracy of 90%.

Yu et al., (2005) have improved the Bayesian

model by applying the tf-idf measure to deal with

the weight of words for Chinese Question

Classification. They have achieved an accuracy of

72.4%. Xu et al., (2006) have employed affiliated

ingredients as the features of the model and used the

results obtained by the syntactic analysis for

extracting the question word. They have also

achieved an accuracy of 86.62% for the coarse

grained categories. Sun et al., (2007) have got

features for classification using HowNet as the

Semantic resource, whereas Yu et al., (2005) have

used Support Vector Machine model, and choose

word, part of speech, chunk, Named Entity, word

meaning, synonyms and Categories coherence word.

Li et al., (2008) have classified open domain

what-type questions into proper semantic categories

using Conditional Random Fields (CRFs). They

have used several syntactic and semantic features

were extracted and compared in their experiments.

They have used the CRFs model to label all the

words in a question, and then choose the label of

head noun as the question category and achieved an

accuracy of 85.60%.

Yu et al., (2007) have presented their

implemented medical QA system, MedQA, which

generates paragraph-level answers from both the

MEDLINE collection and the Web. The system in

its current implementation deals with just

definitional questions (e.g., “What is X?”).

Jacquemart and Zweigenbaum (2003) have

described semantic based approach toward the

development of a French Language Medical QA

System. They have proposed a semantic model of

Medical Questions Classification. In fact, they have

modeled the forms of the 100 Medical Questions as

Syntactic-Semantic Patterns and with one hundred

questions that have been used in their study cannot

cover all QTs. For example, the authors don’t take

into account the What type of questions that

considered one of the most complicate questions to

classify (Li et al., 2008).

Weissenborn et al. (2013) have presented a QA

System for factoid questions in the biomedical

domain. Their system is able to answer only factoid

questions (e.g. “Where in the cell do we find the

protein Cep135?”). The authors have not taken into

account other QTs and have not addressed the

Biomedical QTs Classification problem.

Yang et al. (2015) have described a Biomedical

QA System deals with just factoid questions.

However, the authors have not addressed the

Biomedical QTs Classification challenge. They have

clearly noted that the Biomedical QTs Classification

is a big challenge for building an extensible

Biomedical QA System.

To our knowledge, the literature has not

discussed the QTs Classification problem in the

biomedical domain, whereas our study addresses this

problem of Biomedical QTs Classification and takes

into account all Biomedical QTs in order to build an

extensible Biomedical QA System.

3 PROPOSED METHOD

Our main goal is to classify the Biomedical

Questions into three broad categories: Yes/No,

Factoid and Summary Questions. To achieve this

goal, we propose several Syntactic Patterns of each

Biomedical QTs. We first run a POS Tagger on each

question of Benchmark dataset using the Stanford’s

POS Tagger (Toutanova and Manning, 2000) and

manually analyze these questions and constructing

Syntactic Patterns for QTs. On the basis of that we

have classified each QTs into three categories.

Table 1 show that Which, How and What Types

of questions could belongs to Factoid and Summary

Questions, while Why Type of questions belongs to

only Summary Questions, etc.

Table 1: Question Categories and their Excepted Answer

Types.

Question Category Answer(s)

How

Factoid

Biomedical Entity Names,

Number(s), short expression

Summary

Phrase, Paragraph, short text

summarization

Why Summary

Phrase, Paragraph, short text

summarization

Where Factoid

Biomedical Entity Names,

Number(s), short expression

Which

Factoid

Biomedical Entity Names,

Number(s), short expression

Summary

Phrase, Paragraph, short text

summarization

What

Factoid

Biomedical Entity Names,

Number(s), short expression

Summary

Phrase, Paragraph, short text

summarization

Yes/No Yes/No Yes or No

Biomedical Question Types Classiﬁcation using Syntactic and Rule based Approach

267

To detect Yes/No Questions we used the regular

expression (see pattern (1)), where the questions

should start with three types of words. We found

that this method is significantly efficient.

[Be verbs |Modal verbs |Auxiliary

verbs]+[.*]+?

(1)

Where:

 Be verbs = «am, is, are, been, being, was,

were»

 Modal verbs= «can, could, shall, should, will,

would, may, might»

 Auxiliary verbs= «do, did, does, have, had,

has»

Examples:

Is intense physical activity associated with

longevity?

Does Serca2a bind PLN in the heart?

As it is difficult to distinguish Factoid Type of

Question Wh-word, from Summary Type of

Question Wh-word (Biswas et al., 2014). In this

paper, we have used WordNet in order to improve

the performance of classification. Indeed, The

WordNet (Fellbaum, 1998) is a large English

lexicon in which meaningfully related words are

connected via cognitive synonyms (synsets). The

WordNet is a useful tool for word semantics analysis

and has been widely used in Question Classification

as semantic feature for machine learning approach

such as (Schlaefer et al., 2007).

Additionally, we have used MetaMap (Aronson,

2001) for mapping terms in questions to UMLS in

order to extract the Biomedical Entity Names

(BENs). The UMLS (Bodenreider, 2004) is a

repository of biomedical vocabularies developed by

the US National Library of Medicine.

To overcome this issue, we list all regular

expression patterns of each QTs that are used in our

experiments as follow (NOUN is a noun, JJ is an

Adjective):

What ─ Question category: Summary or Factoid

Question Pattern:

What+[is|are]+

(2)

Examples:

What is the definition of autophagy?

What is the function of the yeast protein Aft1?

What+[is|are]+ (3)

Examples:

What is clathrin?

What is the ubiquitin proteome?

What+[Modal: does]+

Summary»

(4)

What+

(5)

Example:

What types of cancers and inherited diseases have

been associated to mutations in the Notch pathway?

What+[is|are]+

(6)

Examples:

What is the indication of Daonil (Glibenclamide)?

What is the prevalence of short QT syndrome?

Which ─ Question category: Summary or Factoid

Question Pattern:

Which+

(7)

Example:

Which is the phosphorylated residue in the promoter

paused form of RNA polymerase II?

Which+[is|are]+

(8)

Example:

Which is the prevalence of cystic fibrosis in the

human population?

definition

role «Summary»

treatment

aim+[of]+[NOUN|JJ+NOUN]+?

effect

mechanism

Synonyms of those words

Where NOUN is BEN

numbe

name «Factoid»

indication

value

+[of]+[NOUN|JJ+NOUN]+?

frequency

prevalence

frequency

Synonyms of those words

Where NOUN is BEN

[NOUN]+[.*]+?

Or «

Factoid»

[Modal (does| do)] + [.*] + [stand for

bind to

]

numbe

name «Factoid»

indication

value

+[of]+[NOUN|JJ+NOUN]+?

frequency

prevalence

frequency

Synonyms of those words

Where NOUN is BEN

[NOUN]+ ? «Summary»

Where NOUN is BEN

[NOUN]+[.*]+[do]+?

Where NOUN is BEN

[NOUN] where NOUN is BEN

[Verb]+[.*]

+[NOUN]+ [.*]+[NOUN]+?

[JJ+NOUN]

«Factoid»

KDIR 2015 - 7th International Conference on Knowledge Discovery and Information Retrieval

268

Which+[is|are]+

(9)

Example:

Which is the definition of pyknons in DNA?

Where ─ Question category: Always Factoid

Question Pattern:

Where + [.*] +? «Always Factoid» (10)

Example:

Where in the cell do we find the protein Cep135?

Why ─ Question category: Always Factoid

Question Pattern:

Why + [.*] +? «Always Summary» (11)

Examples:

Why does the prodrug amifostine (ethyol) create

hypoxia?

How ─ Question category: Factoid or Summary

Question Pattern:

How + (12)

Example:

How many recombination hotspots have been found

in the yeast genome?

How +

(13)

Example:

How does adrenergic signaling affect thyroid

hormone receptors?

ow are thyroid hormones involved in the

development of diabetic cardiomyopathy?

Define| Synonyms+

(14)

Example:

Define marine metaproteomics.

In addition, we have proposed an algorithm (see

algorithm (1)) to classify the given Biomedical

Question into any one of the predefined categories

using the questions patterns presented above. Then,

the appropriate Answer Extraction Algorithms can

be applied for extracting the precise and most

appropriate answer for that Biomedical Question.

Algorithm 1: Biomedical QTs Classification.

Input: Biomedical Questions

Output: Biomedical QTs /*Yes/No, Factoid or

Summary*/

1: If Biomedical Question matched to [Be verbs |

Modal verbs | Auxiliary verbs] + [.*] +? then

2: Return “Yes/No Question”

3: Else

4: Case 1: Wh-word = “How”

If wh-word+ [Adjective| Adverb] then

5: Return “Factoid”

6: Else if wh-word+ [Modal| Verb] then

Return “Summary”

7: Else if wh-word+[Noun] && Noun =

BEN then Return “Summary”

8: End if

9: Case 2: Wh-word = “Why” Return “Summary”

10: Case 3: Wh-word = “Where” Return “Factoid”

11: Case 4: Wh-word = “Which”

12: If wh-word+ [Noun| Verb|JJ+Noun]&&

Noun = BEN then Return “Factoid”

13: Else if wh-word+ [is| are]+ [[indication of]|

[number of]|…| [synonym of those words]

]+ [NOUN|JJ+NOUN] && Noun= BEN

then Return “Factoid”

14: Else if wh-word+ [is| are] + [[definition

of]| [role of]|…| [synonym of those

words] ]+ [NOUN|JJ+NOUN] &&

Noun= BEN then Return “Summary”

15: End if

16: Case 5: Wh-word = “What”

17: If wh-word+ [is| are] + [Noun] && Noun =

BEN | wh-word+ [Modal: does] + [NOUN]

+ [.*] + [do] +? Then Return “Summary”

18: Else if wh-word+ [Noun] | wh-

word+[Modal]+[.*]+[ stand for| bind to]+?

Then Return “Factoid”

19: Else if wh-word+ [is| are]+ [[indication of]|

[number of]|…| [synonym of those words]

]+ [NOUN|JJ+NOUN] && Noun= BEN

then Return “Factoid”

20: Else if wh-word+ [is| are] + [[definition

of]| [role of]|…| [synonym of those

words] ]+ [NOUN|JJ+NOUN] &&

Noun= BEN then Return “Summary”

21: End if

22: Case 6: Biomedical Question matched to [Define|

Synonyms] + [NOUN] +? && NOUN=BEN

23: Return “Summary”

24: End if

4 EXPERIMENTS AND RESULTS

In this section, we evaluate our proposed method for

definition

role «Summary»

treatment

aim

+[of]+[NOUN|JJ+NOUN]+?

effect

mechanism

Synonyms of those words

Where NOUN is BEN

[Adverb]

Or + [.*] +? «Factoid»

[Adjective]

[Modal]

[Verb] + [.*] +? «Summary»

[NOUN] where NOUN is BEN

[

OUN] +? «Summary»

Where NOUN is BEN

Biomedical Question Types Classiﬁcation using Syntactic and Rule based Approach

269

Biomedical QTs Classification through the study of

the performance of Biomedical Syntactic Patterns

for each type of questions.

4.1 Datasets

The 1433 Biomedical Questions of the Benchmark

datasets are the evaluation text collection of our

study. Those over one thousand Biomedical

Questions were obtained from BioAsk challenges

(Tsatsaronis et al., 2012). Each type of question was

assigned to one category. For example, the question

“What is the role of the Tsix gene during X

chromosome inactivation?” was assigned to

Summary Questions.

Table 2 shows the 3 categories and the number

of Questions Types assigned to each one. For

example, 198 What Type of questions were assigned

to Factoid Questions, 234 What Type of questions

were assigned to Summary Questions, etc.

Table 2: Question Types and their Question Categories.

Question

Types

Question Categories

Yes/No Factoid Summary

Total

How 0 29 37 66

Why 0 2 12 14

Where 0 15 0 15

Which 0 394 55 439

What 0 198 234 432

Yes/No 467 0 0 467

Total 467 638 328 1433

4.2 Results and Discussion

To conduct the experiments, we have used the

Benchmark datasets (Tsatsaronis et al., 2012) that

have been presented above. We have applied

Stanford’s POS Tagger (Toutanova and Manning,

2000) for finding the syntactic structure of

questions. We proceeded by passing the questions

one by one to Stanford’s POS Tagger in order to

capture their syntactic structure. Additionally, We

have applied MetaMap (Aronson, 2001) to extract

the Biomedical Entity Names of each Biomedical

Questions. Moreover, we have used WordNet in

order to generate synonyms of words (e.g. treatment,

effect, etc.) that have been presented in patterns

[2,6,8,9].We have exploited the Syntactic patterns

that have been presented in section 3.

For evaluation, accuracy performance have been

widely used to evaluate Question Types

Classification methods (Li and Roth, 2002; Zhang et

al., 2003; Metzler et al., 2004; Biswas et al., 2014).

Table 3: Accuracy Performance of Biomedical QTs

Questions.

Question

Types

Total Matched Mismatched Accuracy

How 66 63 3 95.45%

Why 14 13 1 92.85%

Where 15 14 1 93.33%

Which 439 434 5 98.86%

What 432 299 133 69.21%

Yes/No 467 467 0 100%

Figure 2: Performance Measurement.

Table 4: Increase performance of what and which types of

questions with/without synonyms.

Question Types What Which

Without Synonyms 59.30% 94.75%

With Synonyms 69.21% 98.86%

Increase Performance +10% +4%

Figure 3: Accuracy performance of what and which types

of questions with/without synonyms.

Overall, from the results presented in Table 3

and Figure 2, it can clearly be seen that the overall

accuracy of our proposed method for Biomedical

QTs Classification is very interesting. The average

performance for automatically assigning a category

to a question was accuracy of 91.62%.

In fact, the proposed method led to the highest

accuracy of 100% and 98.68% for classifying the

Yes/No Biomedical Questions and Which Type of

KDIR 2015 - 7th International Conference on Knowledge Discovery and Information Retrieval

270

questions. As it is difficult to distinguish Factoid

Type from Summary Type of What Type of

questions, nevertheless, our proposed have achieved

an accuracy of 69.21%. Indeed, we can see from

Table 4 and Figure 3 that synonyms extracted from

WordNet of words (e.g. definition, indication, etc.)

that have been presented in section 3 (see pattern (2)

and pattern (6)) enhanced the performance (+10%)

for classifying What Type of questions. In addition,

using synonyms in pattern (8) and pattern (9)

enhanced the performance (+4%) for classifying

Which Type of questions.

We have shown that the proposed method can

classify the Biomedical Questions into three broad

categories: Yes/No, Factoid and Summary

Questions. Indeed those three categories can

summarize all Question Type that can be posed by

user (Tsatsaronis et al., 2012). In other words, these

categories can summarize all possible cases of the

Expected Answer Types. For example: “Which

diseases are caused by mutations in Calsequestrin 2

(CASQ2) gene? “ is a Factoid Questions and the

answer is a Biomedical Named Entity (BNE)

"CPVT", “catecholaminergic polymorphic ventricu-

lar tachycardia”. In addition, one of the most

advantages of our proposed method is that it does

not require a learning phase and therefore it can be

easily integrated in a Biomedical QA System in

order to build an extensible Biomedical QA System.

As mentioned in the Introduction, so far no one

appears to have discussed QTs Classification in the

biomedical domain. Therefore, the importance of

our results using our proposed method thus lies both

in their generality and their relative ease of

application to Biomedical QA System.

In addition, maybe one limitation of our research

was the patterns used for classifying What Type of

question if we take into account its difficulty to

distinguish Factoid Type from Summary Type of

What Type of question. Clearly these patterns that

have been used for classifying What Type question

are not enough to make generalizations about this

one. However, The 69.21% for the What Type of

question is quite low compared to the rest of the

results. The 69.21% stay satisfactory and it could be

improved with more patterns. Indeed the presence of

large numbers of such rules does not introduce any

problems because the rules are represented by a set

of patterns, and Question Types Classification is

conducted by pattern matching.

Remarkably,

the comparative studies that have

been done have focused on QA Systems and due to

the lack of baseline of Biomedical QTs Classification

methods we could not have comparative study.

The main goal of our project is to develop an

extensible Biomedical QA System capable to answer

all Biomedical QTs that is Yes/No, Factoid and

Summary Questions which is composed by three

main components: Questions Processing, Documents

Processing and Answers Processing. Significantly,

we believe that QT Classification may impact

positively or negatively on the reset of the QA

process. Indeed, the output of QTs Classification

will be used to determine the appropriate Answer

Extraction Algorithm.

5 CONCLUSION

In this paper we have presented a novel method for

classifying the Biomedical Questions into three

broad categories that summarize all possible cases of

the Expected Answer Types: Yes/No, Factoid and

Summary Questions. We first defined the syntactic

structure of the Biomedical Questions using

Stanford’s POS Tagger (Toutanova and Manning,

2000) as well as MetaMap for Biomedical Named

Entities Recognition (BNER). We presented the

rules that are represented by a set of patterns for

each Biomedical QTs. Our experimental results

using our proposed algorithm prove that the

Biomedical QTs classification problem can be

solved quite accurately using our proposed method.

The proposed method will allow selecting the

appropriate Answer Extraction Algorithm and

therefore the Biomedical QA System will be able to

treat all types of questions.

In future work we plan to improve the

performance for classifying What Type of questions

by presenting more patterns. We will pay more

attention to integrate our Biomedical QTs

classification system in a Biomedical QA System.

ACKNOWLEDGEMENTS

We thank BioASK challenges for providing us with

Benchmark datasets.

REFERENCES

Aronson, a R., 2001. Effective mapping of biomedical text

to the UMLS Metathesaurus: the MetaMap program.

Proc. AMIA Symp. 17–21.

Athenikos, S.J., Han, H., 2009. Biomedical question

answering : A survey. Comput. Methods Programs

Biomed. 99, 1–24.

Biomedical Question Types Classiﬁcation using Syntactic and Rule based Approach

271

Biswas, P., Sharan, A., Kumar, R., Sciences, S., 2014.

Question Classification Using Syntactic And Rule

Based Approach. Int. Conf. Adv. Comput. Commun.

Informatics 1033–1038.

Bodenreider, O., 2004. The Unified Medical Language

System (UMLS): integrating biomedical terminology.

Nucleic Acids Res. 32, D267–D270.

Ely, J.W., Osheroff, J. a, Ebell, M.H., Bergus, G.R., Levy,

B.T., Chambliss, M.L., Evans, E.R., 1999. Analysis of

questions asked by family doctors regarding patient

care. BMJ 319, 358–361.

Ely, J.W., Osheroff, J. a, Ebell, M.H., Chambliss, M.L., C,

D., Stevermer, J.J., Pifer, E. a, Vinson, D.C., 2002.

Obstacles to answering doctors’ questions about

patient care with evidence: qualitative study. BMJ 324,

1–7.

Ely, J.W., Osheroff, J. a, Gorman, P.N., Ebell, M.H.,

Chambliss, M.L., Pifer, E. a, Stavri, P.Z., 2000. A

taxonomy of generic clinical questions: classification

study. BMJ 321, 429–432.

Fellbaum, C., 1998. WordNet. Wiley Online Library.

Gupta, P., Gupta, V., 2012. A Survey of Text Question

Answering Techniques. Int. J. Comput. Appl. 53, 1–8.

Haris, S.S., Omar, N., 2012. A rule-based approach in

Bloom’s Taxonomy question classification through

natural language processing, in: Computing and

Convergence Technology (ICCCT), 2012 7th

International Conference on. pp. 410–414.

Jacquemart, P., Zweigenbaum, P., 2003. Towards a

medical question-answering system: A feasibility

study. Stud. Health Technol. Inform. 95, 463–468.

Khoury, R., 2011. Question Type Classification Using a

Part-of-Speech Hierarchy. Lect. Notes Comput. Sci.

212–221.

Li, F., Zhang, X., Yuan, J., Zhu, X., 2008. Classifying

what-type questions by head noun tagging. 22nd Int.

Conf.

Li, X., Roth, D., 2002. Learning question classifiers. Proc.

19th Int. Conf. Comput. Linguist. 1 1–7.

Metzler, D., Metzler, D., Croft, W.B., Croft, W.B., 2005.

Analysis of Statistical Question Classi cation for Fact-

based Questions. Inf. Retr. Boston. 8, 1–30.

Niu, Y., Hirst, G., 2004. Analysis of semantic classes in

medical text for question answering, in: Proceedings

of the ACL 2004 Workshop on Question Answering in

Restricted Domains. pp. 54–61.

Prager, J., Radev, D., Brown, E., Coden, A., Samn, V.,

1999. The Use of Predictive Annotation for Question

Answering in TREC8. Proc. 8th Text Retr. Conf. 399–

411.

Schlaefer, N., Ko, J., Betteridge, J., Pathak, M.A., Nyberg,

E., Sautter, G., 2007. Semantic Extensions of the

Ephyra QA System for TREC 2007., in: TREC.

Sun, J., Cai, D., Lv, D., DONG, Y., 2007. HowNet based

Chinese question automatic classification. J. Chinese

Inf. Process. 21, 90–95.

Sung, C.L., Day, M.Y., Yen, H.C., Hsu, W.L., 2008. A

template alignment algorithm for question

classification. IEEE Int. Conf. Intell. Secur.

Informatics, 2008, IEEE ISI 2008 197–199.

Tomuro, N., 2004. Question terminology and

representation for question type classification.

Terminology 10, 153–168.

Toutanova, K., Manning, C.D., 2000. Enriching the

Knowledge Sources Used in a Maximum Entropy

Part-of-Speech Tagger. Proc. Jt. SIGDAT Conf.

Empir. Methods Nat. Lang. Process. Very Large

Corpora 63–70.

Tsatsaronis, G., Schroeder, M., Paliouras, G., Almirantis,

Y., Androutsopoulos, I., Gaussier, E., Gallinari, P.,

Artieres, T., Alvers, M.R., Zschunke, M., others, 2012.

BioASQ: A Challenge on Large-Scale Biomedical

Semantic Indexing and Question Answering. AAAI

Fall Symp. Inf. Retr. Knowl. Discov. Biomed. Text.

Weissenborn, D., Tsatsaronis, G., Schroeder, M., 2013.

Answering Factoid Questions in the Biomedical

Domain. BioASQ@ CLEF 1094.

Xu, W.E.N., ZHANG, Y., Ting, L.I.U., Jin-Shan, M.A.,

others, 2006. Syntactic Structure Parsing Based

Chinese Question Classification [J]. J. Chinese Inf.

Process. 2, 4.

Yang, Z., Gupta, N., Sun, X., Xu, D., Zhang, C., Nyberg,

E., 2015. Learning to Answer Biomedical Factoid and

List Questions OAQA at BioASQ 3B, in: Working

Notes for the Conference and Labs of the Evaluation

Forum (CLEF), Toulouse, France.

Yu, H., 2006. System and methods for automatically

identifying answerable questions.

Yu, H., Lee, M., Kaufman, D., Ely, J., Osheroff, J. a.,

Hripcsak, G., Cimino, J., 2007. Development,

implementation, and a cognitive evaluation of a

definitional question answering system for physicians.

J. Biomed. Inform. 40, 236–251.

Yu, Z., Ting, L., Xu, W., 2005. Modified Bayesian model

based question classification. J. Chinese Inf. Process.

19, 100–105.

Yu, Z.-T., Fan, X.-Z., Guo, J.-Y., 2005. Chinese question

classification based on support vector machine.

Huanan Ligong Daxue Xuebai (Ziran Kexue Ban)/ J.

South China Univ. Technol. Sci. Ed. 33, 25–29.

Zhang, D., Zhang, D., Lee, W.S., Lee, W.S., 2003.

Question classification using Support Vector

Machines. Proc. 26th Annu. Int. ACM SIGIR Conf.

Res. Dev. Inf. Retr. (SIGIR ’03), July 28 - August 1,

2003, Toronto, Canada 26–32.

Zhang, L., Zhou, L., Chen, J.J., 2007. Structure analysis

and computation-based chinese question classification.

Proc. - ALPIT 2007 6th Int. Conf. Adv. Lang. Process.

Web Inf. Technol. 39–44.

KDIR 2015 - 7th International Conference on Knowledge Discovery and Information Retrieval

272