Kansei Engineering Flow

What is Kansei Engineering?

Kansei engineering is a kind of technology that translates the customer’s feeling into design specifications (Nagamachi and Lokman 2010).

The R&D team grasps the customer’s feeling, namely the Kansei; analyzes the Kansei data using psychological, ergonomic, medical, or engineering methods; and designs the new product based on the analyzed information. Kansei/affective engineering is a technological and engineering process from Kansei data to design specifications.

Besides, the collection of adjectives that describe the product or service domain and the potential Kansei needs (i.e., profound needs of the market) called Kansei words (e.g., elegant, masculine, sober, attractive, urbanlike, sexy, heavy). The Kansei words are collected from various sources such as the team of experts, designers, experienced users, advertisements, magazines, ideas, direct observation, and interviews.

Hybrid Kansei Engineering System

There are six types of KE including category classification, Kansei Engineering System (KES), KE modeling, hybrid KE, virtual KE, and collaborative KE. Category classification is a breakdown technique from the targeted concept for a new product or service to the associated subjective Kansei to the objective design parameters. KES comprises databases and inference engine to support a computerized system that handles the process of interpreting consumer’s feeling and emotion as a perceptual design element. KE modeling utilizes mathematical modeling as logic in a computerized system. Hybrid KE is a type of KES by forwarding KES (FKES) and Backward KES (BKES).

The Diagram of Hybrid KES (Nagamachi 2011)

In this study, we propose the Hybrid KES adopted from Matsubara and Nagamachi (1997) as the general framework of KES, which is the combination of FKES and BKES. First, we describe the support for student decision making and show the concept of Hybrid KES. Then, we propose the Kansei inference model which is based on a multilinear regression model. Next, we show the detailed description of the Hybrid KES structure and design recommendation on algoritma teaching. Finally, we construct the recommendation for the Hybrid KES as the domain for the algoritma teaching design and evaluate the rules obtained.

Kansei Engineering Design

In this article, we want to build a teaching design based on Hybrid Kansei Engineering approach. The purpose of this research is to design a pleasurable learning based on students’ preferences in Algoritma Data Science Indonesia. There are several steps to achieve the main objective, the steps are drawn below:

Hybrid Kansei Engineering Flow Chart

Term Frequency- Index Document Frequency

After collecting Kansei Words or opinion from students, we weight the words collected using Term Frequency - Inverse Document Frequency. The purpose of this step is to know words that is important.

Before conducting the text processing using Term Frequency-Inverse Document Frequency (TF-IDF), the data passed several steps in order to normalize the data. The basic steps for normalizing and text preprocessing are (Miner et al. 2012):
- Determine the scope of the text to be processed (documents, paragraphs, and so forth)
- Tokenize: Break text into discrete words called tokens.
- Filter: Remove stop words (“stopping”) or take wordlist
- Stem: Remove prefixes and suffixes to normalize words
- Normalize case: Convert the text to either all lower or all upper case

After the above steps are performed, we then calculate the term frequency-inverse document frequency (TF-IDF). TF-IDF is a statistical method aimed to reflect how important a word is to a document in a collection. It is a way to score the importance of words (for “terms”) based on how frequently they appear (Djatna and Hidayat 2014). To calculate TF-IDF, the following equation is used (Manning 2008):

\(tfidf_t=tf_{t,d}*idf_t\)

where \(tfidf_t\) refers to term \(t\) a weight in document \(d\), tf is the number of occurrences of the term in the document, while \(idf_t\) is calculated with the following equation:

\(idf_t=log N/df_t\)

where \(N\) is a total number of document and \(df_t\) is the number of documents containing the term \(t\). The weighting words for this step is the adjective words obtained from the students’ opinion questionnaire about students’ expectation in teaching.

Opinion Extraction using PCA

Principal Component Analysis (PCA) is one of the feature extraction tools that are powerful in reducing the dimensional data without losing much information (Bouzalmat et al. 2014). In this research , PCA is essentially reducing the dimension of words collected. To deploy PCA in this research, first, define \(X\) as a matrix of words evaluation for p dimensions which denotes as \(X=\{x_1,x_2,…,x_p\}\) Corr where \(x_i∈X\). Then calculate the mean \(μ_i\) of \(X\) for \(n\) teaching design sample \(i=\{1,2,…,n\}\) which denotes as \(M={μ_1,μ_2,…,μ_p }\).

\(μ_i=1/n \sum_{i=1}^nX_i\)

The next step is to calculate the subtract matrix between \(X\) and \(μ\) which denotes as
\(Y=\{y_1,y_2,…,y_p\}\).
\(y_i=x_i-μ_i\)

Afterward, in order to see the correlation between KW, compute the covariance of the matrix of \(Y\) which denotes as \(C\) with p x p dimensions as

\(C=1/((n-1)) Y^T*Y\)

Generate eigenvalue and eigenvector of \(C\) as formulated where eigenvector denotes as
\(B={b_1,b_2,…,b_p}\)

and eigenvalue denotes as

\(λ={a_1,a_2,…,a_p}\).

\(C*B=λ*B\)

According to the result of eigenvalue \(λ\), we determined how many principal components (PC). Generally, the PC which has eigenvalue over one is retained. Then, the group of the words into their PC according to their loading components which are generated from matrix eigenvector \(B\) and the new design concept is then obtained from the interpretation of words within PC obtained.

# prcomp()

The output of this process is the determination of several variables that most contribute to PC1. Hence, those variables(Kansei Words) will be the new design concept of learning in Algoritma.

The Identification of Design Element and It’s Categories

In KE research, service design element has a significant contribution to creating student/person attractiveness (Nagamachi and Lokman 2015). Students will feel attracted when they experience study in a new design of teaching whether it impacts their feelings or not. The main idea is when the students do not feel attracted and feel an impact when they experience the teaching, the opportunity to get touch with the student is not developed, and there are no chance of the teaching gains students’ interest; hence the students’ competence does also not improve. This condition showed that teaching design elements are essential to give the impression to students directly (Muharam 2011). Based on our observation, the design element that exist in Algoritma teaching:

element <- data.frame(No. = c(1,2,3,4),
                      Method = c("Demonstration", "Simulation", "Practical Work", "Lecture"), 
           Tools = c("White Board", "Mentimeter", "Mind map", "Video"),
           Evaluation = c("Quiz", "Learn by Building", "Mentimeter Scale", "Dive deeper"),
          Instructor_background = c("Computer Science", "Statistics", "Mathematics", "Others"))
element

Sample of Algoritma teaching format

Besides, here is a sample of Algoritma teaching format that will be evaluated by our alumni.

Ringkasan kegiatan :
1. Instructor menjelaskan konsep dengan contoh kasus riil
2. Instructor menunjukkan sebuah contoh kasus riil dan memancing keaktifan siswa dengan melempar beberapa pertanyaan berupa open-ended question sehingga imajinasi siswa terbentuk akan suatu konsep tertentu
3. Instructor mengkaitkan kasus riil dengan formula matematika yang dijabarkan di papan tulis
4. Instructor mengevaluasi hasil pengajarannya dengan berdiskusi dengan para siswa di kelas untuk memastikan materi transfer pengetahuan telah terjadi

Semantic Differential Scale

After weighting words by TF-IDF, the words collected will be evaluated by students using Semantic Differential (SD) scale. The SD Scale was introduced by Osgood et al. (1957), he used 7 points SD scales gathering for the evaluation. In this study, the words collected from students were paired with their antonym, then it was structured into 7 points SD scale starts from -3 to 3. This evaluation has been done by scaling the most influenced words from student opinion to the samples of Algoritma teaching format given.

PETUNJUK :
Berilah tanda silang [x] untuk menilai kesesuaian konsep desain pengajaran yang merepresentasikan preferensi anda terhadap layanan pengajaran yang ada. Skala yang digunakan pada penilaian ini adalah skala semantic differential -3 s.d 3 dengan keterangan:

-3 Sangat tidak setuju sekali
-2 Sangat tidak setuju
-1 Tidak setuju
0 Biasa saja / netral
1 Setuju
2 Sangat setuju
3 Sangat setuju sekali

Semantic Differential Questionnaire Example

Quantification Theory Type 1 (QT1)

Quantification theory Type 1 most often has been used to analyze direct and quantitative relationships between a Kansei word and design elements

Analysis of Beer Can Design Using QT1 (Example)

We want to design a product of beer cans based on Kansei Engineering using Quatification Theory Type 1 Method. Based on 56 samples collected, there are 3 design elements that should be exist in designing a beer can, they are Color, Illustration, and Label shape. Besides, each of the design element has several category. The list of design elements and its category presented below.

table1 <- data.frame(Can_Color = c("1: White", "2: Silver", "3: Gold", "4: Blue", "5: Black", "6: Red", "7: Green", "8: Cream", "9: Light Blue", "10: Yellow"),
                     Can_Illustration = c("1: Animal", "2: Bird", "3: Peson", "4: Barley/Hops", "5: Crown/Symbol", "6: Other Object", "7: No Illustration", " ", " ", " "),
                     Label_Shape = c("1: Oval Shape", "2: Other Trad", "3: None of Above", " "," "," "," "," "," "," " ))
table1

In this documentation, we use bitter that describe someone’s kansei/feeling. Hence, we will design a beer can that describe ‘bitter’ customer’s preference. Here are lists of design elements and how bitter the sample.

data <- data.frame(Name_of_Beer = c("Mack", "Koff Red", "Olvi", "Coors Light", "Stroh’s NA", "Carlsberg", "Texas Select", "Cass White", "Schlitz Blue Ox", "Saku", "Murphy's", "Tiger", "Newquay", "Miller", "Swan Light", "Carlsberg Special", "Belgian Gold", "Bass Pale Ale", "Barhu", "Brahma", "Michelob Golden", "Staropramen", "Lowenbrau White", "Piel’s Red", "Malibu", "Whitbread Pale Ale", "Cass Blue", "Kaisordom", "Pig’s Eye", "Heineken", "Cobra", "Dressler", "Red Wolf", "Old Milwaukee", "Labatt Blue", "West End", "Clausthaller", "Buckler", "Young’s London Lager", "Royal Dutch", "Red Bull", "Lapin Kulta", "Karjala", "Belgium Brown", "Old Milwaukee NA", "Heineken Dark", "Tuborg", "Hite", "Budweiser", "Schaefer Light", "Bewry", "Koff Black", "Lowenbrau Blue", "Miller Lite", "Stroh’s Deep", "Coors Gold"),
  Color = c(4,6,5,2,3,1,1,1,2,4,8,4,4,1,1,1,3,5,5,3,3,8,8,1,2,4,4,1,3,7,5,2,6,6,4,3,1,1,10,7,5,4,6,6,6,6,7,8,1,2,1,5,9,6,6,3),
  Illustration = c(1,5,6,5,5,5,4,5,1,5,5,5,6,2,2,6,4,5,1,4,5,4,5,5,6,5,7,5,3,5,1,5,1,5,7,7,5,5,1,5,1,7,5,4,5,5,5,7,5,4,5,5,5,2,5,5),
  Shape = c(1,3,3,3,3,3,3,3,3,1,2,1,2,3,1,3,3,1,3,3,1,3,3,3,3,1,3,1,3,1,3,3,1,3,3,3,3,1,3,1,3,3,1,3,3,1,1,3,3,1,3,3,3,3,3,3),
  Bitter = c(3.25, 3.63, 4.50, 3.13, 4.00, 2.38, 3.00, 2.25, 4.13,
3.38, 4.13, 3.75, 3.50, 3.00, 3.25, 2.50, 2.50, 4.38, 3.88,
3.75, 4.13, 2.00, 2.38, 2.25, 3.13, 4.00, 3.38, 2.88, 4.38,
3.63, 4.50, 2.63, 4.38, 2.75, 2.38, 3.63, 2.63, 2.50, 2.00,
3.13, 4.75, 3.25, 3.50, 4.00, 3.25, 4.25, 4.25, 2.88, 2.88,
3.63, 3.00, 4.38, 2.25, 3.25, 3.75, 3.13))
data[,2:4] <- lapply(data[,2:4], as.factor)
data

The documentation of Quantification theory type 1 can be found in qt1script.R. We will call several functions to get the sugestion of beer can’s design.

source("qt1script.R")
(coba <- qt1(data[,2:4], data[,5]))

## [1] "Color"        "Illustration" "Shape"

##                category score
## Color.1              -0.51975
## Color.2               0.09824
## Color.3               0.32514
## Color.4              -0.21824
## Color.5               1.07885
## Color.6               0.33662
## Color.7               0.09741
## Color.8              -0.57991
## Color.9              -0.84572
## Color.10             -1.41385
## Illustration.1        0.24555
## Illustration.2        0.07370
## Illustration.3        0.88656
## Illustration.4       -0.09869
## Illustration.5       -0.07258
## Illustration.6       -0.16124
## Illustration.7        0.11760
## Shape.1               0.29910
## Shape.2               0.98491
## Shape.3              -0.17777
## constant term         3.34607

summary(data)

##         Name_of_Beer     Color    Illustration Shape      Bitter     
##  Barhu        : 1    1      :12   1: 7         1:16   Min.   :2.000  
##  Bass Pale Ale: 1    6      : 9   2: 3         2: 2   1st Qu.:2.848  
##  Belgian Gold : 1    4      : 8   3: 1         3:38   Median :3.250  
##  Belgium Brown: 1    3      : 7   4: 6                Mean   :3.346  
##  Bewry        : 1    5      : 6   5:30                3rd Qu.:4.000  
##  Brahma       : 1    2      : 5   6: 4                Max.   :4.750  
##  (Other)      :50    (Other): 9   7: 5

PCC shows the most influential mixture composition that affected customer impression

plot(coba)

category score indicates the preference degree. Based on the plot above, we can get the conclusion that Color.5 or Black, Peson, Other Trad.

The weakness of Quantification Theory Type 1 method:
1. In a quantification theory type 1, simultaneous equations could not have been solved when the number of variables exceeds the number of samples, since in Kansei Engieering, many cases have a larger number of design variables than its samples
2. A problem of interactions between x variables; if there are heavy interactions between x variables (Multicolinearity), its analyzing result is distorted

Evaluation using Fisher Exact Test

After getting the new design concept, we will conduct an evaluation using Fisher’s Exact Test to evaluate the design elements generated. Fisher’ exact test is a statistical significance test used in the analysis of cross-tabulation (Fisher 1954). In this research, the Fisher exact test was used to assess the rule productivity of teaching design generated. In the evaluation, the test directly conducted to show whether the rule \(R: f_i\to Y\) is productive or not (Zaki and Meira 2013). Let consider \(f_i\) as a feature or design element generated where \(F = \{f_1, f_2, ..., f_m\}\) and Y refers to student preference which scores above 4. The equation of \(p-value\) calculation based on fisher exact test is represented as in

\(p-value = \frac{(a+b)!(c+d)!(a+c)!(b+d)!}{n! a! b! c! d!}\)

Where:
\(a\) = the value of appearance probability of both \(f_i\) and \(Y\)
\(b\) = the value of appearance probability if \(f_i\) but not \(Y\)
\(c\) = the value of appearance probability of \(Y\) but not \(f_i\)
\(d\) = the value of appearance probability of neither both \(f_i\) nor \(Y\)
\(n\) = the number of records within a dataset

Design <- data.frame(Design_Elements = c("Metode Diskusi", "Evaluasi LBB", "Mentimeter sebagai tools"), Suka = c(" ", " ", " "), Tidak_Suka = c(" ", " ", " "))

Design