Proceedings of the 6th WSEAS International Conference on Simulation, Modelling and Optimization, Lisbon, Portugal, September 22-24, 2006 452
Estimation of FAQ Knowledge by Classifying Questions and Answers
JUN HARADA1, MASAO FUKETA2, KAZUHIRO MORITA2, TORU SUMITOMO2,
ELSAYED ATLAM2 AND JUN-ICHI AOE2
1
Dept. of Intelligent Mechanical Systems Engineering, Kochi University of Technology,
185 Miyanoguchi, Tosayamada-Cho, Kami-Shi, 782-8502, JAPAN
2
Dept. of Information Science and Intelligent Systems, University of Tokushima
2-1 Minami-Josanjima-Cho, Tokushima-Shi, 770-8506, JAPAN
Abstracts:- Question and answering (QA) systems in the CRM scheme require both the user’s satisfaction
related to the quality and the amount of questions to be managed, it depends on the cost. This paper presents an
estimation method of the FAQ service by introducing the following measurements: 1) user’s disrepute for
products which defined by four types of classifying questions (IMPOSSIBLE, SIDE EFFECT, INSUFFICIENT
and UNCLEAR) and the degree for each type is defined; 2) kindness for solutions replied which is defined by
four types of classifying answers (ACTION, CONFIRMATION, EXPLANATION, and NO PROBLEM) and the
degree for each type is defined; 3) sufficiency for the whole FAQ service that introduced by the 1) and 2).
Key-Words:- Question and answering, Frequently Asked Questions, Semantic Expressions.
1. Introduction This paper presents a measurement of quality of
Question and answering (QA) systems in the CRM the FAQ service by introducing the following
scheme require both the quality relating user’s measurements. 1) Measurement of user’s disrepute
satisfaction and the mount of questions to be for products is defined by classifying questions.
managed, that is to say, it depends on the cost. There Questions are classified by four types IMPOSSIBLE,
are many QA researches for large text databases, but SIDE EFFECT, INSUFFICIENT and UNCLEAR,
they are not relation to CRM schemes [1] [2] [3]. and the degree for each type is defined.
Useful for the CRM researches includes answering 2) Measurement of kindness for solutions replied is
opinion questions by [7], good and bad expression defined by classifying answers. Answers are
understanding by [5], sentence subjectivity by [6], classified by four types ACTION, CONFIRMATION,
and estimating sentence types by [4]. EXPLANATION, and NO PROBLEM, and the
One solution is to control the QA systems by degree for each type is defined. 3) Measurements of
human operators like a call center. In fact, operator sufficiency for the whole FAQ service is introduced
can achieve response with good quality for users, but by the 1) and 2). This degree is defined by an integer
it is impossible to manage many questions because of and it becomes very easy to estimate the FAQ service.
high human expenses. A FAQ (Frequently Asked
2. Semantic Expression of FAQ
Questions) scheme is a well known approach that Knowledge Bases
users can find appropriate answers. It is difficult to In the FAQ dialogue, a questioner (a user or a
determine whether the FAQ service is sufficient, or customer) expects that a respondent (a company
not. Hammond et al. [8] presented knowledge person) provides useful answers resolving his/her
navigation of FAQ systems, but there was no claim. This section discusses a formal definition for
discussion about evaluation of the FAQ service. The FAQ dialogue systems by defining a Q-Class
behavior of the FAQ scheme is similar to the QA attribute. The Q-Class attribute means the degree of
system that takes questions as the input and replies questioner’s disrepute and it is defined by four kinds
answers because FAQ knowledge bases are static in of classes. For examples, “Can not print out” means
general. Therefore, it is very important study to Q-CLASS is [IMPOSSIBLE], “printer is noisy”
analyze and to classify questions and answers of FAQ means Q-CLASS is [SIDE EFFECT], “Printing
knowledge. character is unclear” [INSUFFICIENT] and “The red
, Proceedings of the 6th WSEAS International Conference on Simulation, Modelling and Optimization, Lisbon, Portugal, September 22-24, 2006 453
lamp of the printer has been lighted up” means Q-
CLASS is [UNCLEAR]. p = “The printed character is unclear”
We can define the degree of user’s disrepute by Question semantic expression
using Q-CLASS. By introducing additional SEMANTIC(p)
attributes, the formal description of semantic
expressions for questions and answers are as follows:
[Definition 1] [[[printing],[SITUATION]];[character],[OBJECT]];
The question semantic expression SEMANTIC (p) [[unclear],[CLAIM]];[[INSUFFICIENT],[Q-CLASS]]]
for question p is defined as [C(SITUATION);
x(OBJECT); i(CLAIM); a(Q-CLASS)]. Consider
question p =“The printed character is unclear”. Answer semantic expression
The question semantic expression is TRANSFORM(SEMANTIC(p))
[[[printing],[SITUATION]];[[character],[OBJECT]];
[[unclear],[CLAIM]];[[INSUFFICIENT],[Q-CLASS]]].
For the answer semantic expression, a A-CLASS [[[printing],[SITUATION]];[[cartridge],[OBJECT]];
attribute is introduced. It is four values of answers. [[change],[SOLUTION]];[[ACTION],[A-CLASS)]]]
For example, “Change a cartridge” is [ACTION],
“Check a printer cartridge” is [CONFIRMATION],
“Printed characters becomes unclear if there is little Generation answer q in
ink residual quantity in the cartridge” is SURFACE(TRANSFORM(SEMANTIC(p)))
[EXPLANATION] and “It is not a failure” is [NO
PROBLEM]. q = “Please change the cartridge”
[Definition 2]
The answer semantic expression for question Figure 1 Understanding process for question
semantic expression SEMANTIC(q) is defined as
TRANSFORM(SEMANTIC(p)). Figure 1 shows an illustration of understanding
Consider question semantic expression process for the question “The printed characters are
SEMANTIC(p) = unclear” and the answer semantic expression.
[[[printing],[SITUATION]];[[character],[OBJECT]]; In this case, value [printing] of attribute
[[unclear],[CLAIM]];[[INSUFFICIENT],[Q-CLASS]]]. SITUATION is kept as the same value in the answer
It means that [the user hopes that the printed semantic expressions, but there are many semantic
character becomes clear] and one of the expected expressions to be changed. Consider question p’=
answers should be [recommend change the cartridge]. “Paper is got blocked in printing”
The answer semantic expression SEMANTIC(p’) =
TRANSFORM(SEMANTIC(p)) becomes [[[printing],[SITUATION]];[[paper],[OBJECT]];
[[[printing],[SITUATION]];[[cartridge],[OBJECT]]; [[block],[CLAIM]];[[SIDE EFFECT],[Q-CLASS]]]
[[change],[SOLUTION]];[[ACTION],[A-CLASS)]]]. and TRANSFORM(SEMANTIC(p’)) includes
We can define the degree of answers by using A- [[[paper setting],[SITUATION]];
CLASS. [[sheet holder],[OBJECT]];[[fixed],[SOLUTION]];
[Definition 3] [[ACTION],[A-CLASS]]] as one of answer semantic
For the semantic expression r, SURFACE(r) defines expressions. This case means the situation of
a set of surface sentences. question p’ is focusing on the detailed situation in the
Consider the question semantic expression r = answer.
[[[printing],[SITUATION]];[[cartridge],[OBJECT]];
[[change],[SOLUTION]];[[ACTION],[A-CLASS)]]], 3. Estimation Measurements of FAQ
SURFACE(r) includes sentence “The printed
character is unclear”. By the same manner, for the Knowledge Bases
answer semantic expression r = 3.1 Degree of disrepute for questions
[[[printing],[SITUATION]];[[cartridge],[OBJECT]]; In the question understanding process, affective
[[change],[SOLUTION]];[[ACTION],[A-CLASS]]], information (user’s tone, sentence style and so on.)
SURFACE(r) includes “Please change a cartridge”. are considered and Q-CLASS defines in the question