PROPOSING A SIMILARITY MEASURE IN CASE BASED

REASONING FOR PRODUCTS SELECTION

An Experimental Evidence

Fadi Amroush

Department of Economic Theory and History, University of Granada, Granada, Spain

Keywords: Decision Support Systems, Case Based Reasoning, Products Selections.

Abstract: This paper presents a novel similarity measure to design a Decision Support System for products selection

using Case Based Reasoning "CBR". The presented approach combines a novel local similarity measure

with Nearest Neighbour Matching Function which is used as a typical evaluation function to compute the

nearest-neighbour matching case in CBR. This paper suggests using this similarity measure in CBR in order

design our model in products selection to help users to find the optimal product according to their

preferences. The nature of this local similarity measure is to give more reality measure used by people in

selecting products instead of the traditional one proposed by (Xiao-tai et al., 2004). We illustrate the

significance of our proposed measure experimentally. The paper shows that our approach has been followed

by about 80% of subjects.

1 INTRODUCTION AND

LITERATURE REVIEW

The product selection will become more important

nowadays especially as online products can be used

to give online consumers a better choice of products

than can be found in traditional shops. Selecting

what products you'll buy is one of the most common

decisions a consumer makes, mostly the consumer

selects products based on his personal preferences or

hot-item lists consideration. Let us assume that you

want to buy a laptop, and there are many types of

products different in prices and characteristics. The

main question is how are you going to select the best

product -laptop- that meets your preferences within

the budget you allocated, and without paying more

for characteristics you don’t need?

In this work, we focus on proposing a decision

support system to help people in selecting the

optimal product using CBR technique. The concept

of optimality has first appears in 1890, it means

"most favourable", that is meaning to look for the

best possible compromise solution to a problem,

when there are several competing considerations, not

all of which can be simultaneously maximized. The

solution is induced by applying CBR steps through

the process of retrieving the stored cases, calculating

the similarity ratio between these cases and the new

case, and then selecting the most similar case. The

novel idea here is to suggest a new similarity metric

for products selection and test its significance

experimentally.

Using the Case based Reasoning "CBR"

techniques in the decision making process is one of

many methods raised with the appearance of data

mining techniques (Kolodner, 1991). CBR is

implemented in large scale in many arias, Yang et al

(2009) presented in his paper a Case Based

Reasoning Decision Support System (CBR-DSS)

that assists contractors in solving mark up estimation

problem. This proposed CRR-DSS uses successful

cases of previous completed projects to derive

solution to new project mark up estimation problem,

the principle of the CBR-DSS was to analogy new

project with previous projects. Schmitt and

Bergmann (1999) suggested applying CBR

technology for Product Selection and Customization

in Electronic Commerce Environments. Another

study is given by Lin et al (2010) which focused on

strategy selection for product service system design,

in this study CBR is utilized to provide suggestions

for finding this appropriate strategy. Ricci and

Werthner (2002) adapted a case based querying for

travel planning recommendation, it adapts CBR to

499

Amroush F..

PROPOSING A SIMILARITY MEASURE IN CASE BASED REASONING FOR PRODUCTS SELECTION - An Experimental Evidence.

DOI: 10.5220/0003746504990502

In Proceedings of the 4th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2012), pages 499-502

ISBN: 978-989-8425-95-9

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

provide personalized recommendations system

based on previous system experience, and it applies

query refinement methods helping to adjust queries

according to the data available in a given product

catalogues

2 OUR PROPOSED SIMILARITY

MEASURE

In this section we will propose our similarity

measure and how we use it in our model.

2.1 Nearest Neighbour Function

Nearest-neighbour retrieval is one of the most

Nearest-neighbour retrieval is a famous approach

that computes the similarity between stored cases

and new input case based on weight features.

Nearest-neighbour retrieval is a simple approach that

computes the similarity between stored cases and

new input case based on weighted features.

Traditionally, the similarity between queries, Q and

a case, C is defined as the sum of the similarities of

its constituent features multiplied by their relevance

weights, as in the following equation Eq1.

),(

sim

CaseRCaseIsimilarity

∑

(1)

Eq1. Nearest-neighbour evaluations function.

Where w

is the importance weight of a

characteristic, sim is the similarity function and f

and f

are the values for characteristic i in the input

and retrieved cases respectively.

We have started from the typical evaluation function

which is proposed by Kolondner, 1991 in equation

(2). Local similarity Sim is calculated rationally as

following (Xiao-tai et al., 2004):

sim

−

−= 1),(

(2)

Where

is the value scale of the characteristic i,

From our point of view, this function is not

suitable for products selections, because when a

product has a characteristic supported more than it is

needed, it will be considered like than it is not

supported.

In this paper we have enhanced this function to

be more realistic for products selection. We suggest

for product selections if the product has more

support for a specific characteristic so that is not bad

and also should not affect others. Thus the deference

local similarity is not suitable.

we propose The following definition of the local

similarity between two vectors "The ratio of the

minimum level that the consumer wants to have in

his product to the degree of support that the product

has for this characteristic, but without exceeding 1".

Let us assume:

• P: The degree of support a product has for the

characteristic.

• C: The minimum level the consumer wants to

have in the product for the characteristic.

Local Sim can be computed as follows:

If C==0 then Sim = 0 else

If C>P then sim = 1

Else Sim = P/C

(3)

2.2 Example

Let us assume that there are three products P1, P2,

and P3, which have the following main

characteristics A, B, and C. On the other hand there

are given preferences "R" as following, the

minimum level the consumer wants to contain in the

product is as following for the same A, B, and C.

- The characteristic A is: Essential thus, has 10/10.

- The characteristic B is: Desirable thus, has 5/10.

- The characteristic C is: Not important thus, has

0/10.

Now there are the following products P1, P2 and

P3 which have the degree of support of the

characteristics A, B and C as following table2.

Table 1: The degree of support for three products.

A B C

P1 7 5 0

P2 5 10 5

P3 10 1 0

R 10 5 0

2.2.1 Applying Our Proposed Local

Similarity Metric

Applying Eq1, Eq3 for calculating full similarity:

Sim(P1,R)=(0.7)(0.66)+(1)(0.33)+(0)(0)=0.792.

Sim(P2,R)=(0.5)(0.66)+(1)(0.33)+(1)(0)= 0.66.

Sim(P3,R)=(1)(0.66)+(0.2)(0.33)+(0)(0)=0.726.

The sorting is P1, P3, P2, so in our approach P1

is the best option.

ICAART 2012 - International Conference on Agents and Artificial Intelligence

500

2.2.2 Applying Traditional Approach

Now we will Follow the traditional approach by

applying Eq2, Eq3 for calculating full similarity:

Sim(P1,R)=(0.7)(0.66)+(1)(0.33)+(0)(0)=0.792.

Sim(P2,R)=(0.5)(0.66)+(0.5)(0.33)+(1)(0)=0.49

Sim(P3,R)=(1)(0.66)+(0.6)(0.33)+(0)(0)=0.858.

The sorting is P3, P1, P2, so in traditional

approach P3 is the best option.

3 EVALUATION

Finding a concrete manifestation of the term

“similarity” is usually the task of a knowledge

engineer, from the basic knowledge, there are no

experiments to investigate similarity measure for

products selections, However, the last decade has

seen a number of approaches that aim at using

Machine Learning techniques to adjust the similarity

assessment in Case-Based Reasoning.

3.1 The Experiment

An Experiment has been done to investigate how

Subjects select products according to their

preferences of the minimum level they want their

product to contain, and are they follow the

informatics system advice. In the Experiment we

follow the section 3.1.1 in order to compare results.

There are 4 stages in the experiment, in each

stage we asked the subjects to have a decision to

select one product if they want to buy according to

their given preferences. These questions are

including: own opinion, opinion about the others

selection, opinion after proposing our similarity

measure, final decision after proposing a prize of 10

or 20 Euros if the selection is common among the

others, this prize to assure the he will select

seriously what he think is the best, not doing as

lottery. Here is the forth stages, and theirs question

to the subjects.

3.2 Running the Experiment

The experiment consisted of two sessions, in each

session, there were twenty three participants. All of

the sessions were conducted at Laboratory de

Experimental Economic ¨EGEO¨ in the Faculty of

Economic and Business, at University of Granada,

Spain. Spanish is the language of experiment. The

subjects were recruited from undergraduate courses

and some post graduated in business and economics

at the same university. The experiment was

computerized using the Web based program

developed by the author. There were two treatments

differ in the amount of the prize. Each subject

earned 5euros as show up fees and additional prize

(20 Euros or 20 Euros, nothing). In average earned

16.95 Euros.

3.3 Results

The number of subjects who select P1 in stage 4 is

increased in the second treatment, the other factors is

similar, so there is no deference between the two

sessions - treatments-, and there is no significant

factors, we will handle with whole group in addition

to see gender effect, males and females.

As mentioned above, three have 4 Stages, here

figure5-which shows how many subjects select each

product in each stage.

Figure 1: Number of subjects who select each product in

each stage.

It is clear to notice that .number of subjects that

selecting P1 is increasing stage after stage, as

figure3 shows. Here is how the informatics system

information affects decisions of subjects in paid

stage S4.

Figure 2: How many subjects select products at final stage.

PROPOSING A SIMILARITY MEASURE IN CASE BASED REASONING FOR PRODUCTS SELECTION - An

Experimental Evidence

501

3.4 Discussion

The results give an experimental evidence for our

proposed similarity measure which find out that P1

is the best whereas the traditional one finds out that

P3 is the best. About 80% of total subjects agree

with our proposed similarity measure. (63% follow

it regardless of the prize).

There is no age effect, Age is not important for

males: The average of male's age in both with

selecting P1 and without is 21.3, whereas the

average of Female's age is 21.87, and who selected

P1 22,05. The amount of prize is not affecting

subjects in selecting more P1.

4 CONCLUSIONS

Characteristic based product selection well be

famous especial on the Internet, and with integrating

with e-commerce, so e- tailors must provide systems

to support online products selection. Case-based

reasoning is an approach that can provide a solutions

to the problem of Products selection, all based on a

knowledge representation and similarity metric.

In the context of CBR, we present in this paper a

decision support model for products selections, we

have presented a novel local similarity metric for

products selection and compare it with traditional

one. The evidence presented indicates the effective

of our proposed similarity advise, and showed that

subject follow it when it is presented as informatics

advise. An experimental study is conducted to

investigate how people select products, we reported

the results and how subjects change their selections.

ACKNOWLEDGEMENTS

I acknowledge and warmly appreciate the

tremendous support from my supervisors (Nikolaos

Georgantzís, López Herrera, Antonio Gabriel) from

Universidad de Granada, Thanks for their helpful

comments.

REFERENCES

Aamodtn, A., Plaza, E. (1994). Case-based reasoning:

Foundational issues, methodological variations and

system approaches, AICom - Artificial Intelligence

Communications, IOS Press, Vol. 7: 1, (pp. 39-59).

Babka, O., Whar, S. Y. (1997). Case-based reasoning and

decision support systems, Intelligent Processing

Systems, ICIPS '97. IEEE International Conference,

Issue Date: 28-31 Oct 1997, vol.2, (pp. 1532 - 1536).

Breese, J. S., Heckerman, D. (1995). Decision-Theoretic

Case-Based Reasoning, The Proceedings of the 5th

International Workshop on Artificial Intelligence and

Statistics, Fort Lauderdale, FL, January, 1995.

Coyle, L., Doyle, S., Cunningham, P. (2004).

Representing Similarity for CBR in XML, the

proceedings of the 7 the European Conference on

Case Based Reasoning, ECCBR, 2004, PP. 119-127.

Empirical Similarity Itzhak Gilboa, Offer Lieberman,

David Schmeidler, THE PINHAS SAPIR CENTER

FOR DEVELOPMENT, Discussion Paper No.3-2006,

June, 2006.

Gayer, G., Gilboa I. (2007). Offer Lieberman, Rule-Based

and Case-Based Reasoning in Housing Prices, the B.E.

Journal of Theoretical Economics, Volume 7, Issue 1

2007 Article 10.

Gilboa, I., Schmeidler, D. (1995). Case-Based Knowledge

and Planning. Discussion Papers 1127, Northwestern

University, Centre for Mathematical Studies in

Economics and Management Science.

Khoshgoftaar, T. M., Seliya, N ., Sundaresh N. (2006) An

empirical study of predicting software faults with

case-based reasoning, Software Quality Journal - SQJ,

vol. 14, no. 2, pp. 85-111. DOI: 10.1007/s11219-006-

7597-z

Kolodner, J. L. (1991) Improving human decision making

through case based decision aiding, AI Magazine,

12(2), pp .52–68.

Kumar, V., Viswanadham, N. (2007). A CBR-based

Decision Support System Framework for Construction

Supply Chain Risk Management, Proceedings of the

3rd Annual IEEE Conference on Automation Science

and Engineering, Scottsdale, AZ, USA, Sept 22-25,

2007.

Kun, H L., Li, H S., Shing, S L., & Yung T L. (2010).

Strategy selection for product service systems using

case-based reasoning, African Journal of Business

Management, Vol. 4(6), pp. 987-994, June 2010.

McDermott, J. (1980). R1: an expert in the computer

system domain, Proc National Conference on

Artificial Intelligence, (pp.269-71).

Schmitt, S., Bergmann, R. (1999). Applying Case-Based

Reasoning Technology for Product Selection and

Customization in Electronic Commerce Environments,

Global Networked Organizations Twelfth

International Bled Electronic Commerce Conference

Bled, Slovenia, June 7 - 9, 1999.

Turban, E. (1995). Decision Support and Expert Systems:

Fourth Edition, Prentice Hall International, Inc.,

London.

Yang, Z., Deng, F., Liu, W., & Fang, Y. (2009). A CBR

method for CFW prevention and treatment, Expert

Systems with Applications, 36 (2009), pp.5469–5474.

Yin, Z., Li Y. (2010). Intelligent Decision Support System

for Bridge Monitoring. In Proceedings of the 2010

International Conference on Machine Vision and

Human-machine Interface (MVHI '10). IEEE Com-

puter Society, Washington, DC, USA, 491-494. DOI=

10.1109/MVHI.2010.203 - http://dx.doi.org/10. 1109/

MVHI.2010.203.

ICAART 2012 - International Conference on Agents and Artificial Intelligence

502