=Paper=
{{Paper
|id=Vol-1097/STIDS2013_T17
|storemode=property
|title=An Ontological Inference Driven Interactive Voice Recognition System
|pdfUrl=https://ceur-ws.org/Vol-1097/STIDS2013_T17_AbabnehWijesekera.pdf
|volume=Vol-1097
|dblpUrl=https://dblp.org/rec/conf/stids/AbabnehW13
}}
==An Ontological Inference Driven Interactive Voice Recognition System==
https://ceur-ws.org/Vol-1097/STIDS2013_T17_AbabnehWijesekera.pdf
An Ontological Inference Driven Interactive
Voice Response System
Mohammad Ababneh Duminda Wijesekera
Department of Computer Science Department of Computer Science
George Mason University George Mason University
Fairfax, VA, USA Fairfax, VA, USA
mababneh@gmu.edu dwijesek@gmu.edu
Abstract— Someone seeking entry to an access controlled accusations of bias from rejected entrants. Ideally, a
facility or through a border control point may face an in successful interview should accommodate differences in
person interview. Questions that may be asked in such an accents and provide assurance that it is unbiased against
interview may depend on the context and vary in detail. One similar attributes.
of the issues that interviewers face is to ask relevant
questions that would enable them to either accept or reject
entrance. Repeating questions asked at entry point Given the recent success of interactive voice response
interviews may render them useless because most (IVR) systems such as auto attendants, satellite
interviewees may come prepared to answer common navigation, and personal assistants such as Apple’s Siri,
questions. As a solution, we present an interactive voice Google’s Voice, Microsoft’s Speech, we investigated the
response system that can generate a random set of questions possibility of specializing IVR systems for access control
that are contextually relevant, of the appropriate level of such as: Visa interviews, entry point interviews, biometric
difficulty and not repeated in successive question answer enrollment interviews, password reset, etc.
sessions. Furthermore our system will have the ability to
limit the number of questions based on the available time,
Although IVR systems have come a long way in
degree of difficulty of generated questions or the desired
subject concentration. Our solution uses Item Response recognizing human voice, and responding to human
Theory to select questions from a large item bank generated requests as if responses come from another human, most
by inferences over multiple distributed ontologies. of the existing IVR systems are pre-programmed with
questions and their acceptable answers, and consequently
have limited capability in satisfying the Use Case at hand.
Keywords—Ontology; Semantic Web; OWL; Dialogue;
Question Answering; Voice Recognition; IVR; VXML; Access The first minor limitation of current IVR systems
Control Policy; Security; Item Response Theory. comes from the fact that, the human starts and drives the
conversation. The second limitation is that most IVR
I. INTRODUCTION systems have a finite number of pre-programmed
conversations. Therefore the set of questions generated by
such a system are the same for every conversation. This
Physical control points such as human guarded gates,
limitation may expose the set of questions so that aspiring
border control points and visa counters provide entry into
entrants may come with prepared question-answer pairs,
facilities or geographical regions to those that can be
even if the subject matter of the questions may be
admitted legitimately. Legitimacy is usually determined
unfamiliar to them. Consequently, having the ability to
by rules, regulations or policies known to entry control
select questions from a large pool may resolve this
personnel whose duty is to ensure that these policies are
limitation. The third limitation is that when selecting a
enforced while admitting people. In order to do so, they
random set of questions from a large pool, the set of
hold an interview, in which an aspiring entrant is asked a
questions asked may not have the desired overall level of
series of questions, and possibly show some documents
difficulty to challenge the user. Solving this issue is
and demonstrate some knowledge about the contents of
relevant because all aspiring entrants expect to have a fair
the documents or attributes contained in them. Successful
interview. The forth limitation is that questions must be
interviews should have questions that are relevant, of a
able to discriminate between someone that knows the
reasonable level of difficulty (i.e. not too difficult or
subject matter from someone who guesses an answer.
common knowledge) and not to have been asked in prior
interviews for the same purpose without drawing
STIDS 2013 Proceedings Page 125
As a solution we created an ontological inference CAT/IRT reduces the number of questions necessary to
based IVR system that uses item response theory (IRT) to reach a credible estimation of the examinee’s ability by
select the questions [13, 3]. Our system uses the XACML 50%. CAT/IRT can be used to control the number and
language as a base to establish entry policies that consist order of questions to be generated based on examinee’s
of rules to specify the attributes that must be possessed by previous answers [4, 5].
permitted entrants [7]. The IVR system has the
responsibility of determining access by asking questions Our goal in this work is to demonstrate and build an
generated using ontological inferences and IRT. access control system using dialogues of questions and
answers generated from a suitable collection of
In previous work, we introduced a policy-based IVR ontologies. Table I shows a sample dialogue that is
system for use in access control to resources [1]. Later, generated from our research. Our prototype automated
we presented an enhancement that uses IRT to select IVR system can help immigration enforcement at a border
queries from a large set of attributes present in a policy control point making a decision to permit or deny a
[2]. Here we introduce ontology-aided access control person asking for entry. Through a dialogue of questions
system by including questions related to the base and answers, the interviewee will be assigned a numerical
attributes in order to ascertain the interviewee’s score that will then serve as a threshold in the decision
familiarity, and provide a score for the entire set of making process. This score is calculated using IRT, which
answers [8]. We also have the added capability to takes into account the correctness of the user’s responses
generate the succeeding question based on the accuracy of and the weight of the individual questions.
the preceding question. We do so by aligning each The rest of the paper is written as follows. Section II
attribute with an ontology that encodes the subject matter describes an ontological use case, Section III describes
expertise on that attribute and derive facts from these the response theory. Section IV describes the system
ontologies using reasoners to generate questions. We then architecture. Section V describes our implementation.
assign weights to these derivations based on the axioms Section VI is about experimental results and section VII
and rules of derivations used in the proof tree. concludes the paper.
Usually ontologies have a large number of axioms II. Motivating Use Case
and assert even more facts when using reasoners.
Consequently, blindly converting such an axiom base to In this section, we describe an example ontology used
human-machine dialogue would result in very long in our work to generate efficient dialogues of questions
conversations with many disadvantages. The first is that
TABLE I. A SAMPLE DIALOGUE
human users would become frustrated of being subjected
to long machine driven interrogations, and thereby
reducing the usability of the system. The second is that
long conversations take longer time to arrive at an
accept/reject decision, and likely to create long queues at
points of service, such as Airports and guarded doors. In
addition, having a line of people behind one person in
close proximity may leak private information of the
interviewee. Also, others may quickly learn the set of
questions and answers that would get them mistakenly
authorized, thereby gaining unauthorized access.
We use IRT, which provides the basis for selecting
tests from large number of potential questions.
Psychmotricans in social sciences and standardized test
preparation organizations such as the Educational Testing
Services that administer standardized test examinations
like SAT, MCAT, GMAT etc. have developed
methodologies to measure an examinee’s trust or
credibility from answers provided to a series of questions.
In traditional tests, the ability of the examinee is
calculated by adding up the scores of correct answers.
Currently, Computerized Adaptive Testing (CAT) that
relies on IRT has been used to better estimate an
examinee’s ability. It has also been shown that the use of
STIDS 2013 Proceedings Page 126
and answers that are used in assigning a numerical value We use this ontology in our work because it serves as a
to an interviewee’s ability or trust level. good example showing the strength of our system. First, it
shows the possibility of generating valuable questions
Fig. 1 illustrates a class diagram of our under- from asserted or inferred facts. Second, it enables the
development ontology for homeland security. The implementation of the theory under consideration (to be
purpose of this ontology is to collect, organize and infer discussed later in the background section) to generate
information that can help deterring possible attacks, efficient and secure dialogs that are used in: (1) making
enforcing strict entry and enabling faster reach to entry control decisions, (2) assigning numerical values to
suspects. The ontology defines classes, individuals, ability or trust in the shortest time possible and (3) load
properties and relationships using OWL 2 Web Ontology distribution among interviewers and diverting people for
Language (OWL) [9]. The major entities in the ontology further investigation.
are:
x Person: defines humans in general and has The use of ontology in such an application provides
subclasses like; International Student and Friend. many benefits. The most important amongst them is
x Event: defines an event that has a location, date, time reasoning. Using a reasoner we are able to derive facts
and type like terrorist attack from asserted ones. These facts are used to generate
x International Student: is a person who is on an F-1 questions to measure the knowledge or ability level of an
or J-1 Visa type interviewee on a subject under questioning. In IRT, better
x University: defines a university. Some of its current item selection and ability estimation happens when a large
members are MIT and GMU set of items is available to draw questions from. Using
x City: defines a city like Boston ontology, the large number of derivable facts provides us
x Country: defines a country like USA, Russia, with the ability to increase the number of questions, and
Dagestan, Kazakhstan, etc. also control the quality and difficulty of questions.
x State: defines a state like Massachusetts
x Visa: defines visa types like F-1 and J-1 student visas Although there are many reasoners such as FaCT++,
and maybe others. JFact, Pellet, RacerPro, we use HermiT [12] in our work.
Given an OWL file, HermiT can determine whether or not
This ontology represents many kinds of data classes the ontology or an axiom is consistent, identify
and relationships between these major classes and subsumption relationships between classes and deduce
individuals. For example, we define the “Boston other facts. Most reasoners are also able to provide
Marathon Bombing” as a “Terrorist Attack” that explanations of how an inference was reached using the
happened in “Boston”, which is a city in “Massachusetts” predefined axioms or asserted facts.
state. Another fact is that “Dzhokhar Tsarnaev” is an
“Event Character” in the “Boston Marathon Bombing” One such fact derived from asserted ones in our
“Terrorist Attack”. Also we have an “International ontology, is finding the friends that hold a student visa of
Student” who is a friend to “Event Character” in the a person involved in a terrorist attack. To explain this, we
“Boston Marathon Bombing”. have “dzhokhar is friend of Dias”, “Dias is friend of
Azamat”, “Dias has F-1 visa”, “Azamat has a J-1 visa”,
“dzhokhar is an “Event Character” in the “Boston
Marathon Bombing”, “Boston Marathon Bombing” is a
“Terrorist Attack”. Thus we infer (using the HermiT
reasoner) that Azamat and Dias are the friends of the
Boston Bomber and therefore need to be questioned at
any entry point. We use this chain of derivations to
generate specific questions from them.
Reasoners and the explanations that they provide are
very important components in our work to generate
relevant and critical questions from ontology that measure
knowledge and estimate ability from a response in order
to grant access or assign trust. In the example above, the
reasoner provided an explanation of the inference using
11 axioms. We use such a number in defining the
Fig. 1. The Homeland Security Ontology in Protégé difficulty of questions generated from such inferences, as
STIDS 2013 Proceedings Page 127
characterization of what happens when an individual
meets an item, such as an exam or an interview. In IRT,
each person is characterized by a proficiency parameter
that represents his ability, mostly denoted by (T) in
literature. Each item is characterized by a collection of
parameters mainly, its difficulty (b), discrimination (a)
and guessing factor (c). When an examinee answers a
question, IRT uses the examinee’s proficiency level and
the item’s parameters to predict the probability of the
person answering the item correctly. The probability of
answering a question correctly according to IRT in a
three-parameter model is shown in (1), where e is the
constant 2.718, b is the difficulty parameter, a is the
discrimination parameter, c is the guessing value and Tis
the ability level [3].
𝑃 = 𝑐 + (1 − 𝑐) ( ) (1)
Fig. 2. A sample explanation of an inferred axiom in Protégé
using the HermiT reasoner
In IRT, test items are selected to yield the highest
information content about the examinee by presenting
will be explained in section V. Fig. 2 shows the HermiT
items with difficulty parameter values that are closer to
reasoner explanation of our inferred fact.
his ability value. This reduces time by asking fewer and
relevant questions rather wider range ones while
III. BACKGROUND satisfying content considerations such as items or rules
that are critical for a decision of access or scoring.
A. IVR Systems
1) IRT parameter estimation
The main purpose of an IVR system is to interact with
humans using a voice stream. An IVR environment In order to determine the difficulty and discrimination
consists of a markup language to specify voice dialogues, parameters of a test item, IRT uses Bayesian estimates,
a voice recognition engine, a voice browser and auxiliary maximum likelihood estimates or similar methods (MLE)
services that allow a computer to interact with humans [3, 4]. In the original IRT, an experiment is conducted to
using voice and Dual Tone Multi-Frequency (DTMF) estimate these values for each item and at an assumed
tones with a keypad enabling hands-free interactions level of ability for various groups with associated values
between a user and a host machine [13]. Recently, many of IRT parameters using his judgment and experience.
applications such as auto attendant, satellite navigation, Nevertheless, by using our system we can also revise any
and personal assistants such as Apple’s Siri, Google’s initial values for these parameters. We model rule
Voice, Microsoft’s Voice, etc., have started using IVR attributes as test items and rely on the policy
systems. The IVR language we use is VoiceXML, administrator to provide the estimated probabilities.
sometimes abbreviated as VXML [14]. Briefly, Voice
XML is a Voice Markup Language (comparable to 2) IRT ability estimation
HTML in the visual markup languages) developed and
standardized by the W3C’s Voice Browser Working In IRT, responses to questions are dichotomously
Group to create audio dialogues that feature synthesized scored. That is, a correct answer gets a score of “1” and
speech, digitized audio, recognition of spoken and an incorrect answer gets a score of “0”. The list of such
(DTMF) key inputs, recording of spoken input , results consist an item response vector. To estimate the
telephony, and mixed initiative conversations. examinee’s ability, IRT utilizes maximum likelihood
estimates (MLE) using an iterative process involving a
B. Item Response Theory priori value of the ability, the item parameters and the
response vector as shown in (2). Here, 𝜃 is the estimated
IRT, sometimes called latent trait theory is popular ability within iteration s. 𝑎 is the discrimination
among psychometricians for testing individuals, and a parameter of item i, where i=1,2,...,N. 𝑢 is the response
score assigned to an individual in IRT is said to measure of the examine (1/0 for correct/incorrect). 𝑃 𝜃 is the
his latent trait or ability. Mathematically, IRT provides a
STIDS 2013 Proceedings Page 128
probability of correct response from (1). 𝑄 𝜃 is the
probability of incorrect response = 1- 𝑃 𝜃 [3,4].
[ ]
𝜃 = 𝜃 + (2)
Then, the ability estimate is adjusted to improve the
computed probabilities with the examinee’s responses to
items. This process is repeated until the MLE adjustment
becomes small enough so that the change becomes
negligible. IRT accommodates multiple stopping criteria
such as: fixed number of questions, ability threshold or a
standard error confidence level. The result is then
considered an estimate of the examinee’s ability and the
estimation procedure stops. The ability or trait usually
ranges from -∞ to +∞, but for computational reasons
acceptable values are limited to the range [-3, +3]. Fig. 3. Ontology-based IVR using IRT
C. Access Control and XACML into VoiceXML and plays to the user. Then the system
waits for the user’s utterance, and if the user provides one,
the system’s voice recognition software attempts to
Access control policies specify which subjects may
recognize the input and checks the correctness of the
access which resources under some specified conditions
answer. Based on the answer, the IRT estimation
[6]. An attribute-based access control policy specifies
procedure either increases a priori ability score or
subjects, objects and resources using some attributes. decreases it. The process continues until a predetermined
XACML is an OASIS standard XML-based language for level of ability or accuracy specified according to the
specifying access control policies [7]. In a typical application is reached.
XACML usage scenario, a subject that seeks access to a
resource submits a query through an entity called a Policy Because ontologies produce a large number of facts, it
Enforcement Point (PEP), which is responsible for would be impractical to run a dialogue that lasts hours in
controlling access to the resource. It forms a request in the order to estimate user’s ability. In our homeland security
XACML request language format and sends it to the a ontology uses 167 axioms. The reasoner was able to infer
policy decision point (PDP), which in turn, evaluates the 94 facts raising the total number of axioms and candidate
request and sends back one of the following responses: to generate questions to 273.
accept, reject, error, or unable to evaluate.
We use IRT to manage and control dialogue questions
IV. USING IRT TO MANAGE AND CONTROL generated from a large pool of ontologically derived facts
DIALOGUES FROM ONTOLOGIES in a way that shortens the length of dialogues while
keeping the maximum accuracy in estimating the user’s
Fig. 3 shows the overall architecture of our system. trust. The IRT-based estimated (θ) represents the trust or
We use derived or axiomatic facts of the ontology to confidence of the system in the person answering the
create questions asked by our IVR system. Given that a questions in order to make an access decision.
large number of facts can be derived from our ontology,
but only few questions can be asked during an interview, We have used the OWL annotation property to assign
we use IRT to select the facts that are used to generate IRT parameters to axioms. Annotations were selected in
questions. order to keep the semantics of the original ontology and
structure intact. We annotate every asserted axiom in the
Our questions are automatically created without ontology with IRT parameters, which are: difficulty (b),
human involvement by combing English words or phrases discrimination (a) and guessing (c). Currently, we assume
such as “Does” or “Is-a” with ones chosen from the all asserted axioms have the same default degree of
ontology of (subject, property, object) triples. The difficulty and discrimination values of 1. The code
expectation is a dichotomous answer of either (yes, no) or snippet in Fig. 4 illustrates our annotation using Java with
(true, false). The ontological property names such as “is- OWL API. An improvement to this approach would be to
a”, “has-something” are prime candidates for creating assign different values for difficulty and discrimination by
true/false questions. Our system transforms the question using domain experts.
STIDS 2013 Proceedings Page 129
OWLAnnotationProperty irtDifficultyAP = Set
df.getOWL inferredAxioms=inferredOntology.getAxioms();
AnnotationProperty(IRI.create("#irt_difficulty" DefaultExplanationGenerator explanationGenerator
)); =new DefaultExplanationGenerator(
OWLAnnotation irtAnnotation = manager, factory, ontology, reasoner, new
df.getOWLAnnotation( SilentExplanationProgressMonitor());
irtDifficultyAP , df.getOWLLiteral(1.0)); for (OWLAxiom axiom : inferredAxioms) {
for (OWLAxiom axiom : axioms) { Set explanation =
OWLAxiom axiom2 = axiom.getAnnotatedAxiom explanationGenerator.getExplanation(axiom);
(Collections.singleton(irtAnnotation)); //Annotate inferred axioms using the number of
manager.addAxiom(ontology, axiom2); explanation
} OWLAxiom tempAxiom =
axiom.getAnnotatedAxiom(Collections.singleton(irt
Fig. 4. Java code for asserted axiom annotation Annotation));
manager.addAxiom(inferredOntology, tempAxiom);
We weigh inferred facts more during the estimation
Fig. 5. Java code for inferred axiom annotation
process. We are calculating these parameter values from
the number of explanation axioms used in each
individually inferred fact. Our current scheme of The resultant decision is based on the IRT
difficulty value assignment is shown in Table II; where characteristics of the axiom and not on the number or the
higher values or weights are assigned according to the percentage of correctly answered questions as in
number of explanation axioms used to infer a fact, and traditional testing. The ability estimate produced by our
consequently the question generated from it is considered implementation also comes with a standard error (SE)
to be more difficult than one generated from an asserted value that is a measure of the accuracy of the estimate.
fact. Fig. 5 illustrates a code snippet for inferred axiom Equation (3) presents the formula used for standard error
annotation. calculation [7].
In our current work and for testing purposes we use a 𝑆𝐸 𝜃 =
∑ ( )
default value of “1.0” for discrimination and “0.0” for
guessing, which practically neutralizes them leaving the
difficulty parameter as the sole factor in estimating ability Higher standard error indicates that the estimate is not
using equation 2. However, our solution and algorithm are very accurate, while lower values indicate higher
based on the IRT two-parameter model, which relies on confidence in the estimation. This too can be used as a
the item’s difficulty and discrimination parameters. Fig. 6 means to discontinue the dialogue or use an alternate
shows our algorithm to estimate ability based on equation decision method.
2 [3]. Our system estimates the ability of a user after
every answer to a question generated from an axiom V. IMPLEMENTING THE ONTOLOGY-BASED IVR
before selecting and asking the next question. If the SYSTEM FOR ENTRY CONTROL
ability estimate exceeds the threshold then access is
granted. If the threshold is not reached then additional Here, we present a prototype of our system showing
questions are offered. If the estimated ability doesn’t the major components. It is not yet validated as a
reach the threshold the dialog stops and access is denied. deployable system, but it works for the sample use case.
Depending on the application, the dialog might be run
Algorithm 1: IRT Ability estimation
again giving a second chance. When the ability estimation
Input:a priori theta, Difficulty, Discrimination,
again reaches a predefined threshold, the system Answer
concludes the dialog and conveys the decision. Output: posteriori theta, standard error
/* calculate theta and standard error*/
TABLE II. IRT DIFFICULTY ASSIGNMENT BASED ON 1:for (counter < items.length) do
NUMBER OF AXIOMS IN EXPLANATION 2: itemDifficulty=parseFloat(difficultyArray[i]);
Number of IRT 3:itemDiscrimination=parseFloat(discriminationArr
ay[i]);
explanations Difficulty 4:answer=parseFloat(answerArray[i]);
1 0 Easy 5:probTheta=calculateProbability(itemDiscriminati
2-3 1 on,aTheta,itemDifficulty); // equation 1
4-5 1.5 Moderate 6:thetaSplus1= claculateTheta(probTheta, thetaS);
//equation 2
6-7 2 7:endfor;
8-9 2.5 8:estimatedTheta = thetaSplus1;
>=10 3 Hard 9:return thetaSplus1;
Fig. 6. Algorithm for ability estimation in IRT
STIDS 2013 Proceedings Page 130
1) Voice Platform (Voxeo)
Voxeo’s Prophecy is a comprehensive IVR and Fig. 7. A sample Homeland security VoiceXML greeting form
standards-based platform [15]. Some of the capabilities
integrated into the platform are: automatic speech The conversation starts with a menu in VoiceXML
recognition, speech synthesis (Text-to-Speech), Software hosted on the local Voxeo Prophecy web server. The
Implemented Phone (SIP) browser and libraries to create voice browser connects to the web server and converts
and deploy IVR or VoIP applications using VXML text to speech and speech to text. Fig. 7 shows a sample
CCXML. It supports most of server side languages and VoiceXML code.
has a built-in web server.
Fig. 8 shows our algorithm integrating ontology, IVR
2) Item bank and IRT. This algorithm was successfully implemented
using JavaScript and Java Server Pages (JSP) embedded
In our work, we start with ontology, annotate every in VoiceXML pages. The main steps are as follows:
axiom with an “irt_difficulty” property of value “1”. Then x Load the ontology and parse the XML into Document
we use this ontology in the HermiT reasoner to infer Object Model (DOM).
implicit axioms and their explanations. The inferred facts x Extract the axiom’s triplet (subject, property, object)
are themselves annotated with “irt_difficulty” property x Extract the axiom’s IRT difficulty value from the
and values calculated by factoring the number of annotation
explanation axioms using the schema stated in Table II. x Establish a VoiceXML “For” loop that synthesizes a
question from string or text values to speech (TTS).
For example, when annotating the inferred fact “the The question consists of an auxiliary verb, object,
friends of the Boston Attack Bomber”, which has an property and subject to test the correctness of an
explanation that includes 11 axioms shown in Fig. 2, the axiom.
irt_difficulty annotation would be “3.0”;; which is the x The system waits for a response. If there is one it
highest value on the scale of IRT difficulty parameter converts it to text and recognizes it. If it adheres to
values in Table II. We assume that answering a question grammar then a value is assigned as an answer.
generated from a high-valued fact is a difficult task.
x If there was no answer then VXML re-prompts the
Consequently, if the answer to a question derived from
question up to a programmed number of times. If
this fact is correct, the ability estimate would be impacted
exceeded then an appropriate VXML is executed.
more positively than a correct, but easy one and more
x The vector of binary answers is used to estimate the
negatively if the opposite happens. An example is the
IRT ability.
asserted axiom that “Boston is located in Massachusetts”.
Because this is an asserted fact, it is annotated with value x The loop continues until a threshold of T or the
“1.0”;; which makes a question generated from it an easy maximum number of questions is reached.
one and thus not affecting the ability estimate greatly. x The IRT ability estimation algorithm, as illustrated in
This process is basically generating the item bank in Fig. 6, takes the variables: answer vector, a priori T,
CAT/IRT terminology. Each item in the item bank difficulty, discrimination and calculates a posteriori
contains a question, an answer and IRT parameters. In 𝜃.
addition to saving it as ontology in any of the supported x If the answer is correct (“yes” or “true”), a value of
formats, this item bank can also be supported by using a “1” is assigned. If not, a “0” is assigned.
more specialized CAT/IRT platform like Cambridge x The last posteriori 𝜃 in the loop is the estimated
University’s Concerto [16]. user’s ability T and can be compared to a threshold
value set by an administrator. Access is granted if ( T
3) Generating dialogues from an ontology > 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑) and denied otherwise.
STIDS 2013 Proceedings Page 131
voice dialogs from ontologies for entry control. We have
Algorithm 2: dialogue access evaluation
Input: a priori theta, Difficulty, used IRT to generate shorter dialogues between the
Discrimination, Answer system and a human speaker. IRT is useful in
Output: access control decision compensating for inaccurate voice recognition of answers
/* make access control decision from
ontology*/
during dialogs or accidental mistakes. Our entry control
1: domDocument=parse(ontology); // DOM decisions are made based on an estimation of a level of
2: subjectArray=getAxiomSubject(axiom); trust in a subject derived from the importance or
3: propertyArray=getAxiomProperty(axiom); relevance of axioms in ontology. The use of IRT also
4: objectArray=getAxiomObject(axiom);
5: difficultyArray=getAxiomDifficulty(axiom);
enables the reordering of questions with the purpose of
6: /*use voiceXML , JSP to generate dialog*/ preserving privacy in IVR systems. With the advancement
7: for (counter < items.length) do in the fields of mobile, cloud and cloud based voice
8: ‘[auxiliary verb]’ recognition such systems become important in defence
+propertyArray[i] + “ ” + objectArray[i]
+“ ”+ subjectArray[i]; and physical security applications [17, 18, 19].
9: = user_utterance;
10: response[i] = REFERENCES
Field.voiceRecognition(user_utterance);
11: if response[i]= ‘Yes’ or ‘true’
[1] M. Ababneh, D. Wijesekera, J. B. Michael, “A Policy-based
12: resultVector[i]=1;
Dialogue System for Physical Access Control”, The 7th STIDS
13: else
2012), Fairfax, VA, October 24-25, 2012.
14: resultVector[i]=0;
15: endfor; [2] M. Ababneh, D. Wijesekera, “Dynamically Generating Policy
Compliant Dialogues for Physical Access Control”, CENTERIS
16: theta = IRT_algorithm(resultVector,
difficulty, discrimination,aPrioriTheta); 2013 - Conference on Enterprise Information Systems – aligning
17: if theta > thetaThreshold technology, organizations and people, Lisbon, Portugal. October
18: permit; 23-25, 2013.
19: else [3] F. B. Baker, The basics of item response theory, ERIC
20: deny; Clearinghouse on Assessment and Evaluation, 2001.
[4] D. J. Weiss, G. G. Kingsbury, Application of computerized
Fig. 8. Ontology-IVR algorithm with IRT adaptive testing to educational problems, Journal of Educational
Measurement, 21, 361-375, 1984,
[5] H. Wainer, Computerized Adaptive Testing: A Primer, Second
Edition, Lawrence Erlbaum Associates Publishers, 2000
VI. EXPERIMENTAL RESULTS [6] M. Bishop, Computer Security: Art and Science, Addison Wesley,
2002.
Our implementation shows that efficient dialogs could [7] XACML, OASIS, URL: https://www.oasis-
open.org/committees/tc_home.php?wgabbrev= xacml, accessed
be generated from ontologies that have been enhanced September 30, 2013.
with IRT attributes. The successful implementation of the [8] T. R. Gruber. A translation approach to portable
IRT in dialogues of questions and answers shortens the ontologies. Knowledge Acquisition, 5(2):199-220, 1993.
number of questions necessary to reach an accurate [9] W3C, Web Ontology Language (Primer),
http://www.w3.org/TR/owl2-primer/ , accessed August 22, 2013.
estimation of subject’s ability, knowledge or trust by at [10] W3C, SPARQL Protocol and RDF Query Language, URL:
least 50% as it has already been proved by the IRT http://www.w3.org/2009/sparql/, accessed August 22, 2013.
literature [4, 5]. This reduction of the number of questions [11] http://owlapi.sourceforge.net/reasoners.html, accessed August 22,
necessary to estimate the ability produces shorter dialogs 2013.
[12] http://hermit-reasoner.com/, accessed August 22, 2013.
without losing accuracy. Also, the use of IRT enables the [13] W3C Voice Browser Working Group, URL:
use of multiple stopping criteria such as: fixed length http://www.w3.org/Voice, accessed August 22, 2013.
number of questions or time, ability threshold and [14] W3C, Voice Extensible Markup Language (VoiceXML)(VXML),
standard error confidence interval. The availability of URL: http://www.w3.org/Voice/, accessed August. 22, 2013.
[15] Voxeo web site, URL: http://www.Voxeo.com, accessed August
large number of ontology axioms enables generating a set 22, 2013.
of questions different from another set to be generated [16] Concerto IRT Platform, URL: http://www.psychometrics.
immediately after the current user preserving privacy and cam.ac.uk/page/338/concerto-testing-platform, accessed August
protecting against question exposure, especially in voice 22, 2013.
[17] Microsoft Windows Phone Speech, URL:
systems. The success of dialog system depends upon http://www.windowsphone.com/en-us/how-to/wp7/basics/use-
multiple timing factors and scalability of supporting speech-on-my-phone, accessed September 3,2013.
multiple users. Our on-going research addresses these two [18] Apple Siri, URL: http://www.apple.com/ios/siri, accessed
aspects. September 3, 2013.
[19] Google Android Mobile Search, URL:
http://www.google.com/mobile/search/, accessed September 3,
VII. CONCLUSION 2013.
We have designed and implemented a novel IVR
system that can dynamically generate efficient interactive
STIDS 2013 Proceedings Page 132