International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016

ISSN 2229-5518
55

The Question Answering System

Using NLP and AI

Shivani Singh Nishtha Das Rachel Michael Dr. Poonam Tanwar

Student, SCS, Student, SCS, Student, SCS, Associate Prof. , SCS
Lingaya’s Lingaya’s Lingaya’s Lingaya’s University,
University,Faridabad University,Faridabad University,Faridabad Faridabad

Abstract:
(ii) QA response with specific answer to a specific question
The Paper aims at an intelligent learning system that will take instead of a list of documents.
a text file as an input and gain knowledge from the given
text. Thus using this knowledge our system will try to answer
questions queried to it by the user. The main goal of the 1.1. APPROCHES in QA
Question Answering system (QAS) is to encourage research

IJSER
into systems that return answers because ample number of There are three major approaches to Question Answering
users prefer direct answers, and bring benefits of large-scale Systems: Linguistic Approach, Statistical Approach and
evaluation to QA task.
Pattern Matching Approach.
Keywords:
A. Linguistic Approach
Question Answering System (QAS), Artificial Intelligence
This approach understands natural language texts, linguistic
(AI), Natural Language Processing (NLP)
techniques such as tokenization, POS tagging and parsing.[1]
These are applied to reconstruct questions into a correct
query that extracts the relevant answers from a structured
1. INTRODUCTION database. The questions handled by this approach are of
Factoid type and have a deep semantic understanding.
Question Answering (QA) is a research area that combines
research from different fields, with a common subject, which B. Statistical Approach
are Information Retrieval (IR), Information Extraction (IE)
and Natural Language Processing (NLP). Actually, current Availability of huge amount of data on the World Wide Web
search engine just do “document retrieval”, i.e. given some increased the importance of Statistical Approach. Statistical
keywords it only returns the relevant ranked documents that approaches and online text repositories are not dependent on
contain these keywords. They do not provide a precise structured query languages and can formulate queries in
answer to that. Hence QAS is designed to help people find natural language.
specific answers to specific questions in restricted domain.
Statistical techniques: Support Vector Machine classifier,
QA systems are classified into two main categories, namely Bayesian classifiers, etc.
open-domain QA systems and closed- domain QA systems.
Open-domain question answering deals with questions about
nearly everything such as the World Wide Web. On the other C. Pattern Matching Approach
hand, closed-domain question answering deals with questions
under a specific domain (music, weather forecasting etc.) The This approach deals with expressive power of text pattern, it
domain specific QA system considers heavy use of natural replaces the sophisticated processing involved in other
language processing systems. computing approaches. This approach uses the
communicative power of text patterns. This approach best
QA is different from the search engines in two aspects: suits to small and medium sized websites. The type of
(i) In QA, query is the question not a keyword questions handled by this approach are mainly factoid based,

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016
ISSN 2229-5518
56

definition based and it has a less semantic understanding as

compared to other approaches. Most of the pattern matching Documentation and
QA systems uses the surface text pattern. Passage Retrieval

Question
1.1. QAS COMPONENTS Processing Document
Question

The architecture of the QA system as mentioned earlier, Answering

would consist of following three modules: 1. Question Query/Questi Processing
processing module. 2. Document processing module. 3. on
Classification
Answer extraction and formulation module.

The questions that the system receives can be divided into Fig.1 Architecture of QAS Answer
two major categories: FACTUAL & EXPERT. Factual
questions are those which contain words like what, where,
when, who, etc. Expert questions are those which contain
words like how, why etc. 2. LITERATURE SURVEY
 TheDocument Processing Module QA systems, as explained before, have a backbone consisting
of three main parts: categorization of question, information
It takes in the choice of the user for a particular passage from retrieval, and answer extraction. Therefore, each of these
the displayed list .Then using POS Tagger tags the tokens

IJSER
three components attracted the attention of QA researchers.
.With the help of tags finds the verbs in the passage. Using
the list of irregular verbs and the logic for regular verbs a Question Classification
data structure (array) is created which contains the verbs
along with their tenses and ing form. Question Question Answer Type
class
 TheQuestion Processing Module: WHAT basic-what/ Money/ No./
what-who /what- Definition/ Title/
It takes in a question from the user Using StringTokenizer when/ what- NNP/ Undefined
tokenizes it and stores it in another data structure returns it where
for further use in the program WHO Person /
HOW basic-how Manner
how-manyhow- Number
 The Question-Answering Module long Time/Distance
how-much how- Money / Price how-
First finds the verb in the question. Matches the verb just muchhow-far much Undefined
found with the tokens created in the document processing how-tall Distance Number
stage. According to the selected case for type of factual how-rich Undefined
question (what, when, etc) it further tries to extract and how-large
formulate the answer.
WHERE Location
The user is first asked to select the passage of his choice and
WHEN Date
then the type of question. The Question processing module
WHICH which-who Person Location
will process the question and pass it to the Question which-where Date
Answering module which will make use of the various which-when NNP
extractions received from the Document Processing phase, which-what
along with the Processed Documents containing the tagged NAME name-who name- Person/ORG.
format of the original input document. By applying required where Location
algorithms this module will pass it to the Formulation module name-what Title / NNP
WHY Reason
for getting the desired answer.
WHOM Person

Questions usually gurantee to predictable language patterns,

and therefore are categorized based on taxonomies.
Taxonomies are distinguished into two main types: flat and

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016
ISSN 2229-5518
57

hierarchical taxonomies. Flat taxonomies have only one level BING is Microsoft’s answer to google and it was launched in
of classes without having sub-classes, on t,he other hand 2009. Bing is the default search engine in Microsoft’s web
hierarchical taxonomies have multi-level classes. Lehnert [2] Browser.it is available in 40 languages. It provides different
proposed “QUALM”. services including image, web and vedio search along with
maps.
QA system used a flat taxonomy with seventeen classes e.g.
PERSON, PLACE, DATE, NUMBER, DEFINITION, Stoyanchev et al (StoQA) , 2008: Contribution In their
ORGANIZATION, DESCRIPTION, ABBREVIATION, research, they presented a document retrieval experiment on a
KNOWNFOR, RATE, LENGTH, MONEY, REASON, question answering system. They used exact phrases, as
DURATION, PURPOSE, NOMINAL,OTHER. constituents to search queries. The process of extracting
phrases was performed with the aid of named-entity (NE)
Zhang and Lee [3] compared various choices for machine recognition, stop-word lists, and parts-of-speech taggers.
learning classifiers using the hierarchical taxonomy propose
such as: Support Vector Machines (SVM), Nearest Neighbors Wolfram|Alpha is a Computational Knowledge Engine that
(NN), Naïve Bayes (NB), Decision Trees (DT). introduces a fundamentally new way to get knowledge and
answers, not by searching the Web sites, but by dynamic
Information Retrieval computations based on a vast collection of built-in data,
algorithms, and methods. [7]. Wolfram|Alpha returns an
Evaluate the use of named entities and of noun, verb, and answer in a form of a table where information, which is
prepositional phrases as exact match phrases in a document relevant to a query, is separated by categories (e.g. an answer
retrieval query. Gaizauskas, and Humphreys [4] described an to a query about some person usually contains such se image,
approach to question answering timeline, notable facts, familial relationships and others).

IJSER
Answer Extraction Answerbag is question answering website where users can
get answers to their questions, whether they're looking for
Finding the answers by exploiting surface text information facts, opinions or simply entertainment. Questions are
using manually constructed surface patterns. In order to answered by Answerbag professional researchers and
enhance the poor recall of the manual hand-crafting patterns, community members.[8]so Answerbag can also be
many researchers gained text patterns automatically such as considered as an expert community question answering
Xu et al [5]. website..

3. PROPOSED WORK Blurtit is an online question answering community that

provides the answers users are looking for and gives them
Nowadays there are many different Question Answering free, 24/7 access to a whole world of information, and to
Systems. Most popular QAS and their main characteristics millions of knowledgeable friends. Blurtit knowledge base
are described below. contains facts, information and users’ opinions, enlarging
with each new answer.
ELIZA
Kangavaril, 2008: Contribution The research presented a
One of the earliest QA systems was ELIZA, developed in model for improving QA systems by query reformulation and
1964. One successful ELIZA application was DOCTOR, a answer validation. The model depends on previously asked
computer program that basically interacted with users questions along with the user feedback.
through a text chat interface, answering questions and Experimental environment and results: The system worked
responding to the users dialog in a way that mimicked the on a closed aerologic domain for forecasting weather
client-centered psychotherapy between a client (the user) and information.
their therapist. Limitations:the model was tested in a sparse experimental
environment in which the domain was very particular and the
Evi (originally known as TrueKnowledge) is a knowledge number of questions relatively small. Also, depending only
Web search engine that assists people gaining what they on the users as a single source for validating answers is a
desire and require through Evi understanding of each user two-edged weapon.
and the world they live in. [6].
Ask.com (originally known as Ask Jeeves) is a question
QUORA is available in English and Spanish launched in
answering-focused web search engine. The original idea
2010. The QUORA community includes some well know
behind Ask Jeeves was to allow users to obtain answers to
people, such as Dustin Moskovitz, Jimmy Wales.
questions posed on daily basis, natural language, as well as

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016
ISSN 2229-5518
58

by traditional keyword searching. The current Ask.com still and analyzing information; can naturally be
supports this, with added support for math, dictionary, and produced as QA queries.
conversion questions. This system tries to “understand” any  Can be used to make online lectures more old-
users query and gives three forms of answer at once: a direct school type by allowing lectures to proceed only
answer, a list of links to webpages on related topics and a list when questions related to the previous lecture are
of similar questions with answers from other question answered correctly.
answering websites.

The existing system can be integrated with a search engine to

The general characteristics of described QAS are summarized enhance the performance.
in Table 1.

Table 1
CONCLUSION
QA System Domain Description Year
ELIZA Closed Attempt to mimic 1964
This paper describe about the Question Answering
basic human
interaction Q&A System for an English Language i.e. it receives query
exchanges. from the user and selects most appropriate answer. QAS
EVI Open Specializes in 2007 is approach to find the correct answer to the question
knowledge base & asked from user. This paper also describes different
semantic search QAS approaches, different types of QAS.QA system
Quora Open Knowledge based 2009

IJSER
help in improving system interaction. In this paper we
answering ability
also concentrated on finding the solution of some
Bing Open Provide diff. 2009
services like image, problem: Answer is restricted to a precise domain, user
web and video has to follow a particular path while entering a question
seach and Extracting correct answer. The solution consists:
Stoyanchev Closed Extract phases 2008 semantic representation for Natural Language, effective
Wolfram/Alp Closed Computation search 2009 logic is to be performed on them and developing a
ha engine formalism to represent the answer verification and
Answerbag Open Web based QA 2003 specific answer extraction. Thus there is great potential
system
for exploring the challenges in QA domain.
Blurtit Open People ask query of 2006
regular user provide
answer based on ACKNOWLEDGMENT
their opinion.
Kangavari Closed Depend on 2008 We sincerely thank prof. PoonamTanwar, Computer
Previously asked department, Lingaya’s University, Faridabad for her valuable
Question. guidance and motivation for this work.
ASK.com Open Web based QA 1996

4. APPLICATION AND FUTURE

SCOPE
QA system handles a question in natural language and tries to
provide its answer in a single statement. With certain
improvements; the proposed system can be used for the
following applications:

 Adding speech recognition abilities to the current

system will enable people with reading disabilities
to take advantage of this system.
 Query processing in database systems and many
analytical tasks that involve collecting, correlating REFERENCES

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016
ISSN 2229-5518
59

[1]. Saranya R, Christopher Augustine, ” Schemes and

Approaches in Question Answering System”, International
Journal of Advanced Research Trends in Engineering and
Technology (IJARTET) ,Vol. II, Special Issue X, March
2015.
[2]. Lehnert, W., 1977. The Process of Question Answering -
A Computer Simulation of Cognition. Yale University, ISBN
0- 470-26485-3.
[3]. Zhang, D. & Lee, W., 2003. Question Classification
using Support Vector Machines.In Proceedings of the 26th
Annual International ACM SIGIR Conference.
[4]. Gaizauskas, R. & Humphreys, K., 2000. A Combined
IR/NLP Approach to Question Answering Against Large
Text Collections.In Proceedings of the 6th Content-based
Multimedia Information Access (RIAO-2000).
[5]. Xu, J., Licuanan, A. &Weischedel R., 2003. TREC2003
QA at BBN: Answering Definitional Questions. In
Proceedings of the 12th Text Retrieval Conference.
[6]. The story of Evi .http://www.evi.com/about/.
[7]. http://www.wolframalpha.com/about.html.
[8]. About us, Answerbag2014.

IJSER
http://www.answerbag.com/about-us/

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 7, Issue 12, December-2016
ISSN 2229-5518
60

IJSER

IJSER © 2016
http://www.ijser.org