A Characterization of Student’s Viewpoint to Learning and its

Application to Learning Assistance Framework

Toshiro Minami

, Yoko Ohura

and Kensuke Baba

Kyushu Institute of Information Sciences, Dazaifu, Fukuoka, Japan

Fujitsu Laboratories, Kawasaki, Japan

Keywords:

Text Mining, Text Analysis, Term-usage, Educational Data Mining, Lecture Data.

Abstract:

Due to the advancement of popularization of university education, it becomes more and more necessary for

university staff to help students by enhancing their motivations to learn in addition to training study skills.

We approach to this problem from lecture data analytics. We have been investigating students’ answer to a

term-end retrospective questionnaire, and found students’ attitude in learning and their academic performance

correlate signiﬁcantly. On the basis of this ﬁnding, in this paper, we propose a framework for assisting students

to improve their learning attitude. It consists of four participants; lecturer, assisting staff including librarian,

data analysts, and learning assistance system built on top of learning management system. We discuss how the

results of our previous studies can be utilized to assist students in this framework. Further, we introduce two

indexes for measuring the weights of a student viewpoint between lecture and themselves, and between good

points and bad points. These indexes show how a student’s viewpoint to the class is located in comparison

with other students’ viewpoints.

1 INTRODUCTION

Due to the popularization of university education, it is

often pointed out that not only the knowledge but also

the learning abilities of students has been decreasing.

In order to deal with such situation, universities have

been paying a great amount of efforts. Most univer-

sities in Japan, for example, set up the faculty devel-

opment (FD) programs and force the professors to at-

tend and try to raise the professors’ educational abil-

ity. They also introduce remedial courses for students

who need to learn preparatory materials, and enhance

the courses for the ﬁrst year students to get used to the

style of teaching in universities. However, students’

academic skills do not seem to improve accordingly.

According to our observation of how students

learn in universities, the main problem of this issue is

not on professor/lecturer’s side such as teaching skill,

class management, or something, but on student’s side

such as diligence, motivation, eagerness, and other at-

titudes to learning. Thus, enhancement of students’

attitudes to learning is inevitable in order to achieve

high academic performance of students.

Considering the varieties of students, we take an

approach based on data analytics, which consists of

two steps: (1) to make a student’s learner model

mainly from lecture-related data, so that the model

includes attitudes to learning by proposing new con-

cepts and measuring indexes for them, and get tips for

the students how to learn and the tips for lecturers how

to teach, and (2) to advise each student according to

his or her learner model as well as advising lecturers

and students as a whole.

To proceed such an approach, we propose a

framework for assisting students with better academic

achievement. Enhancing student’s attitude to learning

is a very important function of this framework. We

also discuss in what way data analytics relate with

the framework for better assistance to students’ learn-

ing. Our approach has an advantage in terms of un-

derstandability of humans. We prefer to choose the

understandable method rather than applying the es-

tablished and more sophisticated methods that are less

understandable for us.

As a part of this approach, we have been ana-

lyzing the answer texts of a term-end questionnaire,

which asked the students to evaluate themselves and

the lectures/lecturer by retrospectively looking back

the class (Minami and Ohura, 2013a). Such data are

considered to be appropriate to analyze the students’

attitudes to the lectures. In the previous studies, we

have found that the students with high examination

Minami, T., Ohura, Y. and Baba, K.

A Characterization of Student’s Viewpoint to Learning and its Application to Learning Assistance Framework.

DOI: 10.5220/0006389706190630

In Proceedings of the 9th International Conference on Computer Supported Education (CSEDU 2017) - Volume 1, pages 619-630

ISBN: 978-989-758-239-4

619

scores use the words which indicate their wide point

of view. By contrast, the students with low grades use

the words closely related to the main topics of lecture.

Such studies of educational data analysis have

been conducted in the research ﬁeld of Educational

Data Mining (EDM) (Romero and Ventura, 2007).

For example, Romero et al. (C. Romero and Her-

vas, 2008) gave a comparative study of data mining

algorithms for classifying students using data from e-

learning system. Its major interest is on predicting the

student’s outcome. Our focus is on the student’s psy-

chological tendency in learning, such as eagerness,

diligence, seriousness. Many studies in EDM use the

target data which are obtained from learning manage-

ment systems (LMSx). By contrast, we intend to ob-

tain our target data in everyday lectures.

Goda et al. (K. Goda and Mine, 2013) proposed a

method of text analysis, where texts are provided by

students as the reports in the everyday lectures, which

consists of three components: previous, current, and

next. Our data are different from them. They come

from homework, exercise, and term-end examination,

together with term-end examination, which are able

to obtain in any ordinary courses. Another difference

of our approach is that we do not set the estimation of

the student’s achievement as the main aim. Our major

aim is to know the student’s attitude to the lectures

and their seriousness to learning.

Ames et al. (Ames and Archer, 1988) studied in

the similar motivation to ours. They investigated the

students’ attitudes to the class and learning by ana-

lyzing the answers to questionnaire items. However,

their underlying data were obtained by asking the stu-

dents to choose the rate from 1 to 5 for each question

item. In our case, even though 2 of our question items

are asking to rate from 0 to 100, other questions are

asking to write the students’ own thought in a free-

text format.

Our data analysis style is also different from the

major studies in EDM. Most of them somehow intend

to analyze the big data, and the data obtained auto-

matically as log data. By contrast, we would rather

take the approach of dealing with small data, because

our target data themselves may be very small (Minami

and Ohura, 2012b)(Minami and Ohura, 2013a). Also,

the data we deal with are somewhat representing hu-

man students, and we, as the staff in an educationalor-

ganization like university, have to educate all of them.

Thus, we have to take attention to all the data as well,

even if they are located in the far-away areas from the

central area, because they represent one or more stu-

dents.

The rest of the paper is organized as follows.

In Section 2, we describe an outline of the main

concept of Learning Assistance Framework (LAF),

which provides the students with learning assisting

service. In Section 3, we show some of our results

in our previous studies, and discuss how they could

be reﬂected in LAF. In Section 4, we deﬁne two in-

dexes which measure the term-usage of the students,

and investigate how the students’ viewpoints are char-

acterized on the basis of the results in Section 3. Fi-

nally, in Section 5, we conclude the discussions and

ﬁndings in this paper and present our future direction.

2 LEARNING ASSISTANCE

FRAMEWORK

In this section, we describe the concept of learning

assistance framework (LAF). LAF should be con-

structed so that it helps the lecturers know better about

their students, as well as it helps the students get more

appropriate advice and enjoy better learning environ-

ment, and thus have better academic performance.

Figure 1 shows how an LAF works. The ﬁgure is

separated into two main areas, that is, the students in

the left area and LAF in the right area. LAF consists

of 4 major components, or participants: lecturer, sup-

porting staff (SS), learning assistant system (LAS),

and data analyst (DA). LAS can be developed as an

extension to a learning management system (LMS)

(e.g., (Moodle, a; Blackboard)) by adding an advi-

sory information system (AIS) as is illustrated in the

ﬁgure. LAS may be developed separately from LMS

if it is more convenient for the university.

The concept of LMS are already used popularly in

many educational organizations, such as universities.

Typical LMS has the functionalities such as manage-

ment of class members, allowing the lecturer to pro-

vide course materials to the students, students’ sub-

mission of homeworks, and assignment/examination

setting and scoring, and many more.

In the framework proposed in this paper, the AIS

part is added on top of the fundamental LMS core

functions, as is shown in Fig. 1. For example, Moodle

provides with the facility for adding plugins (Moo-

dle, b), with which the user can extend the materials

dealing with the system, such as special testing func-

tions and streaming videos. The AIS part has its own

database (AISDB), which keeps the data used for ad-

vising students in learning as well as the data provid-

ing the lecturers with the information about the stu-

dents, which does not be provided in ordinary LMSs.

The lecturer and a librarian are working as the

front-end advising staff. In the ﬁgure, the lecturer not

only gives lectures to the students, but also work as

the main adviser of students, for the lecture, for other

A2E 2017 - Special Session on Analytics in Educational Environments

620

Learning

AssistantSystem

AIS

AISDB

Lecturer

Lecture

Advice

Learnin

Material

LectureData

Student/LectureͲ

relatedInfo.

LearningAssistantFramework

LMS

LMSDB

AISDB

Student

Report

Communication

Consultation

Data

Consultation

Advice

DataAnalyst

DataMining

Other

Students

Supporting

Staff

(Librarian)

Student/LectureͲ

relatedInfo.

Figure 1: Outline of Learning Assistance Framework (LAF).

courses, and for learning in general. It is possible

to assign the roles of lecturer and adviser to different

staff. Then, the lecturer can be more concentrated to

provide good lectures and the adviser can give good

advice to students by spending more time in under-

standing and analyzing the status of the students.

The lecturer uses the functions of LAS by provid-

ing lecture data (by LMS) and getting information re-

lating to students (by AIS). The student-related infor-

mation helps the lecturer with advising each student

as well as all the students of the courses the lecturer is

responsible for. The lecture-related information will

be helpful in recognizing about his/her lectures.

Librarians play an important role as a major sup-

porting staff in this framework. Their role is to assist

the lecturers by providing students with consultation

service, and advice the students such as what learning

material is appropriate, how to study, how to do their

homework. In order to play such a role, librarians

will access to the LAS for getting information/data

about students and lectures. The librarians are sup-

posed to provide the LAS with consultation data as

well. They make record data for consultations, which

will be used as the case data in the latter consultations

as well as those to help the lecturers and students.

The data analyst (DA) located in the right-most

area of the ﬁgure also plays an important role as back-

end adviser for students. Different from LMS, AIS

deals with other types data such as about attitudes,

behaviour, and something, which are more subjective

than those of LMS. Thus, the data and algorithms for

AIS should be maintained regularly, and DA is re-

sponsible for it.

DA’s job includes maintaining AISDB, such as

collecting, updating, integrating data; data analysis

for extracting appropriate information about students’

characters, attitudes, interests, and others; and com-

munication with other participants; lecturers and SSs.

By setting DA as a different staff, the lecturer is able

to concentrate more on the front-end jobs such as the

lectures and advising students.

There are some lecturers who have sufﬁcient

knowledge and skill in data analytics. In such cases, it

should be better for the lecturers to analyze their data

by themselves. Then, the DA helps them by provid-

ing with the analysis results of other classes so that

the lecturers can compare their classes with others,

and they can recognize the relative positions of their

classes.

A lecturer is also able to capture attitude of his/her

student not only in his/her class, but also in other

classes by using information provided by the DA.

Such information should be very useful in order to

deliver good lectures and to advise students for the

lecturer.

Security issue is very important in LAF because

the data dealt with the system and other human par-

ticipants are private data of students. One possible

way to cope with this problem is using renumbered

IDs instead of using the students’ IDs, and does not

use the students’ names and other privacy data by the

supporting staff and the DA, and only the lecturers are

allowed to know the students’ original IDs and names

because they have to evaluate the students and thus

they need such private information.

In the ﬁgure, the DA is represented as a person.

Some university has Institutional Research (IR) divi-

sion which is responsible for data analytics and the

staff records various types of data in the university

including the data relating to students such as aca-

demic performance, consultation records. The staff

in IR may be able to play the role of DA. In such a

A Characterization of Student’s Viewpoint to Learning and its Application to Learning Assistance Framework

621

case, DA staff may be allowed to deal with the orig-

inal data concerning students. There are a variety of

options for the university, from setting up a division

for the role of DA to setting up no DA who works as

expert for data analysis.

The students located in the left-most area are main

users, or customers, of LAF. The one at the left-top

area represents the student who takes a lecture of the

lecturer and study under the support from LAS. The

student will communicate with other students indi-

rectly by using LAS as well as directly by talking,

exchanging messages, and other means.

In the rest part of this paper, we will discuss the

issues such as what kinds of data for AIS are useful,

how to change the questionnaire we have been using

in order to create more suitable data, and other issues.

3 FINDINGS IN OUR ANALYSIS

AND THEIR APPLICATION TO

AIS

In this section, we present some ﬁndings in our pre-

vious studies for lecture data, and show what sort of

information relating to student’s attitude is obtained.

We also discuss how these ﬁndings could be used in

the framework of LAF. Firstly in Section 3.1, we de-

scribe the target lecture data we use for analysis. The

target data were obtained as answers to a term-end

questionnaire which asked the students to evaluate the

course and the students themselves by looking back

what they did in the course. Some questions asked to

answer by number and others asked to answer in free

text. In Section 3.2, we show the ﬁndings in investi-

gating the correlation between the numerical answers

and the students’ examination scores. In Section 3.3,

we deal with the free-text data, and investigate the

correspondence between the students and the terms

they used. In Section 3.4, we investigate the answer

texts from contrasting questions for evaluation, that

is, lectures vs. students themselves, and good points

vs. bad points.

3.1 Target Data

The data used in this paper came from a course in

2009 named “Exercise for Information Retrieval” in

a junior (2 year) college. The students were in year

2 and are going to graduate. The number of regis-

tered students was 35. The course was compulsory

for librarian certiﬁcate. Thus, the students of this

course were more motivated than other courses. The

major aim of the course for the students was to be-

come expert information searchers so that they had

enough knowledge about information retrieval, and

also had enough skills in ﬁnding appropriate search

engine sites and search keywords by understanding

the aim and the background of the retrieval. The

course consists of 15 lectures.

Also, homework was assigned at every lecture.

Its aim was to make the students review what they

had learned in the lecture and to study preliminary

knowledge for the next lecture. At the same time, the

students were requested to write a lecture note every

time, which also aimed to force the students review

what they had learned. The homework score shall re-

ﬂect the frequency and quality of the submitted home-

work.

The term-end examination of the course consisted

of 3 questions. The aim of these questions was to

evaluate the skills on information retrieval, including

the skills for planning and summarizing. These skills

are supposed to have learned and trained in the course,

through their exercises in the classes and while they

do homework. We consider the score of term-end ex-

amination as a measure for student’s academic perfor-

mance.

We also asked the students to answer the ques-

tions as the overall evaluation of them for the course.

The questionnaire we deal with in this paper asked 12

questions. The questions stared with asking to eval-

uate the course: (Q1) what they have learned in the

course, and if they are useful, (Q2) what are the good

points of the lectures, (Q3) what are the bad points

that should be corrected, (Q4) score the course as a

whole, with the numbers from 0 to 100, where the

pass level is 60 as in the same way to the examination

score,

Then, the questions asked to evaluate the student

herself: (Q6) what are the good points of the stu-

dent herself regarding learning attitudes and efforts in

learning during the course period, (Q7) what are the

bad points that should be corrected of the student her-

self, (Q11) score the student herself by considering

her efforts and attitude in the course, with the num-

bers from 0 to 100 as in the same way in (Q4).

The amount of data used in this paper is very

small. Therefore, it is impossible to extract useful in-

formation which is applicable in other classes. Our

main aim of analysis of these small data is to ﬁnd

new analysis methods as many as possible as the ﬁrst

step to data analysis of lecture data we use in this pa-

per. Then, we would apply the methods found in the

ﬁrst step to the data obtained from other classes. The

methods are evaluated according to their applicability

to other classes, usefulness of the extracted informa-

tion for lecturers in advising students.

A2E 2017 - Special Session on Analytics in Educational Environments

622

3.2 Analytics for Numerical Answers

In this subsection, we show our ﬁnding in the previ-

ous studies (Minami and Ohura, 2012a; Minami and

Ohura, 2012b; Minami and Ohura, 2013b; Minami

and Ohura, 2013a) of numerical data regarding the

answers to the questionnaire.

We started with investigating the correlation be-

tween the self-evaluation scores (which is obtained

from (Q11)) and the examination scores. The result

shows that the students who have high examination

scores evaluate themselves from a very low scores up

to a very high ones, which means that those students

who evaluate low would have the self-image that “I

am the person who can do better than what I have

been doing”. These students have a good desire of

self-improvement.

By contrast, the students who have poor perfor-

mance seem to believe in themselves without evi-

dence, and evaluate themselves something like, “I do

fairly well in my study”. Another possibility is that

they actually recognize very well about their poor ef-

forts and poor performance. Still, or maybe because

of it, they wanted to believe that they were not very

poor in their efforts, instead of admitting their poor

efforts. In this way, they could avoid facing what they

really were, and kept their prides. As a result of such a

phenomenon, the correlation coefﬁcient between the

self-evaluation scores and the examination scores be-

comes a negative value of −0.1.

Considering the phenomenon we found in this

study, one possible service for AIS is asking the stu-

dents to evaluate themselves and the lectures from

time to time, and monitor their evaluation values. The

system puts a mark on a student who evaluates with

extremely high or low values, so that the lecturer can

recognize it easily. Lecturer can encourage the stu-

dent if his/her self-evaluation score is too low, and ad-

vise the student to correct their attitude to the course if

his/her self-evaluation score is too high in comparison

with his/her efforts.

3.3 Analytics of Word Usage

In this subsection, we show some of the results we

obtained in our previous analytics of word usage of

students in the answer text to the question (Q1) (Mi-

nami and Ohura, 2014; Minami and Ohura, 2015b;

Minami and Ohura, 2015a).

As we see the words that appear frequently in the

texts, we recognize that the words related to the lec-

tures appear in high frequencies. For example, the

word “Search” appears 88 times in the answers for

(Q1), which is the most frequently used one among

all words. Also, the words “Information” and “Li-

brary” appear in the list. The lecture-related words are

6 (20%) among 30 words, whereas 4 (29%) among 14

words with frequencies more than 10.

In a correspondence analysis between the students

and the terms they used, we divided the students into

5 groups. The member of the group with the highest

average examination score characteristically used the

technical terms and the terms from broader points of

view, in comparing Japan and the world such as “For-

eign”, “National”, and “Japan”. It is interesting to see

that the terms which are relating to the homework as-

signments do not appear in this group. Thus, we can

say that the students in this group attended the lectures

with the attitude of learning in a broad perspective.

Contrastingly, the students in the group with the

lowest average examination score used quite a lot of

frequently-used general terms, and did not use techni-

cal terms at all. It is interesting to see that many stu-

dents used a lot of terms they have learned during the

lectures, e.g., “Learn”, “Master”, “Study”, “Useful”,

and “Use”. Thus, the students in this group look very

diligent and eager to learn superﬁcially, however they

are not. Presumably, they took too much attentions

to the terms themselves which are closely related to

the main topics of the course, and did not pay much

attention to their background, their relation to other

concepts, and their values in our social life.

One possible service for AIS is to monitor the

terms the students use in answers to occasional ques-

tionnaires, and advise the student when his/her view-

points to the topics in the course seem to be too dif-

ferent from other students.

For example, let us suppose students are asked to

answer the question about the terms relating to the

course. If the terms used by a student are extremely

different from the ones expected by the lecturer, AIS

puts a mark on the student, and the lecturer starts con-

sidering what are wrong with the student. By refer-

encing other information about the attitudes and be-

haviour of the student, the lecturer can advise the stu-

dent in an appropriate way.

3.4 Analytics of Terms in Answer-texts

to the Contrasting Questions

This subsection deals with analytics of terms in the

answer-textsfor the questions asking from contrasting

points of view. The question (Q2) asked the students

what are the good point of the lectures, LG in short,

and (Q3) asked the bad points of the lectures that need

to be improved, LB in short. Similarly, (Q6) asked

the good points of the student herself (SG), and (Q7)

asked the bad points that need to be improved (SB).

A Characterization of Student’s Viewpoint to Learning and its Application to Learning Assistance Framework

623

We would like to investigate what kinds of terms

are used in which kinds of evaluation questions, for

lecture, self/student, good point, bad point, and try

to ﬁnd out the students’ viewpoints in these evalua-

tions. We proceed the analysis according to our pre-

vious study (T. Minami and Baba, 2017).

Let n be the number of students (n = 35 in our

case), and let S = {s

, s

, . . . , s

} be the set of stu-

dents. Each student s

(i = 1, 2, . . . , n) answers to

the questions (Q2), (Q3), (Q6), and (Q7). Let Q =

{LG, LB, SG, SB} be the set of questions, and let

Ans

i,q

be the answer text (string of characters) of

the students s

∈ S for the question q ∈ Q. Note that

Ans

i,q

=“” means that the student s

did not answer to

the question q.

By applying the morphological analyzer, i.e., KH

coder (Higuchi) and MeCab (Kudo), to the text

Ans

i,q

, we are able to create the set of “terms”,

, t

, . . . , t

i, j

}, where each term t

is of the form

w ∶ p, where w is a word and p is its part of speech

(PoS). We will sometimes identify the term w∶ p with

the word w in this paper.

Let T

i,q

be the set of terms obtained from Ans

i,q

and #

i,q

t be the number of occurrences, or frequen-

cies, of the term t in the text Ans

i,q

. Note #

i,q

t repre-

sents the number of the occurrences of the term t in

the bag of words of Ans

i,q

, and thus, #

i,q

t = 0 if t ∉ T

i,q

We also deﬁne T

⋃

q∈Q

i,q

, #

t =

∑

q∈Q

i,q

t, T

⋃

∈S

i,q

, and #

t =

∑

∈S

i,q

t. Then, let T =

⋃

q∈Q

or =

⋃

∈S

Now we extend Q = {LG, LB, SG, SB} to Q =

{LG, LB, SG, SB, L, S, G, B, All} so that T

= T

⋃

and #

t = #

t + #

t. We also deﬁne T

, T

, and T

in the same way. Further, T

All

= T

⋃

and #

All

t =

t + #

t. We may omit the sufﬁx ALL sometimes for

brevity.

In our case #T = 605,

∑

t∈T

#t = 1322, and thus, a

word appears about 2.2 times in average. The most

frequently appearing term is the verb “do” with 72

times, and 361 (about 60%) terms appear only once.

In order to investigate how terms are used in con-

trasting answer-texts, we introduce an index, which

quantiﬁes how much is a term used comparativelybe-

tween two texts. Let t be a term (∈ T). The LS-index

of t is deﬁned as follows:

(t) =

t −#

t +#

(1)

By deﬁnition, −1 ≤ ι

(t) ≤ 1, and ι

(t) = 1 iff t ap-

pears only in L, i.e., t appears either one of LG or LB

and it does not appear SG nor SB. Also, ι

(t) = −1

iff t appears only in S, and ι

(t) = 0 iff t appears in

the same number in L as in S, or #

t = #

t. Similarly,

we deﬁne:

(t) =

t −#

t +#

(2)

/*b

/%b

/b*

/b*b

/b1

/b%b

1*b

/b%

6b*b

6b*

6b1

6b%b

6b%

6*b

1%b

6%b

/6,QGH[

*%,QGH[

Figure 2: Distribution of Terms with LS (x-axis) and GB

(y-axis) Indexes.

Table 1: Frequencies for Combined Types of LS and GB.

S S’ N L’ L Sum

G 70 4 10 9 158 251

G’

1 6 2 36 13 58

N 10 2 16 3 18 49

B’

4 16 0 18 0 38

88 3 9 6 103 209

Sum 173 31 37 72 292 605

Figure 2 shows how terms are located 2-

dimensionally between LS and GB indexes. We di-

vide the terms into 25 groups by combining 5 groups

both for LS (x-axis) and for GB (y-axis), namely, S,

S’, N, L’, and L for LS, and G, G’, N, B’, and B

for GB. Precisely, we deﬁne the groups as follows:

S = {t ∈ T ∣ ι

(t) = −1}, S

′

= {t ∈ T ∣ −1 < ι

(t) < 0},

N = {t ∈ T ∣ ι

(t) = 0}, L

′

= {t ∈ T ∣ 0 < ι

(t) < 1},

and L = {t ∈ T ∣ ι

(t) = 1}. We deﬁne G to B in

a similar way, and ﬁnally we deﬁne from SG to LB

by combining the two group types. For example,

′

= {t ∈ T ∣ −1 < ι

(t) < 0, 0 < ι

(t) < 1}.

Even though it is easy to see how terms are dis-

tributed, Figure 2 is misleading because one point

may represent a lot of terms with the same LS and

GB index values. For example, the point located at

the right-top corner, which represents the terms with

the value 1 for both LS and GB index, represents 158

terms; which is the maximum among 25 types.

Table 1 shows the actual numbers of terms for

each type. From the table, we can see most (nearly

70%) terms are located at the 4 corners (namely LG,

SG, SB, and LB types), and #LG> #LB> #SB> #SG in

their numbers of terms. Note that the number of terms

A2E 2017 - Special Session on Analytics in Educational Environments

624

of a type also indicates the amount of viewpoints, and

thus, how widely it is evaluated from.

These results show that students use more terms

regarding (probably, pay more attention to) lectures

than students themselves. Further, they use more

terms, or pay more attention, to good points than bad

points for lectures, and they pay more attention to bad

points than good points for themselves.

A possible interpretation of these results is that the

students are generally generous to others and they try

harder to ﬁnd goodpoints than bad points as they eval-

uate the lectures and the lecturer,and at the same time,

they try hard to ﬁnd something to be improved as they

evaluate themselves. It should be very interesting to

investigate further on this issue.

Regarding the possible application(s) in LAS, the

analysis method itself is important. By applying the

method, we obtain the characteristic feature of a term

between two contrasting concepts, or viewpoints.

4 ANALYTICS OF TERM-USAGE

OF STUDENTS REGARDING LS

AND GB

In this section, we analyze how the terms are used

by students on the basis of the studies described in

the previous subsections as the next step toward un-

derstanding students’ viewpoint. We start with deﬁn-

ing the indexes for a student, which show what sort

of terms she used in regard with the lecture-student

(LS) and good-bad (GB) points of view. The indexes

are deﬁned by using the LS and GB index values for

terms.

Two approaches are possible in order to capture

students’ attitudes to learning. One is a direct way;

by asking questions about their attitude, for example.

Our approach is an indirect way; by asking questions

about different types of questions and trying to ﬁnd

their attitudes by analyzing answers of students. We

are able to capture how students recognize about their

own attitude to learning from the direct approach,

whereas we may be able to capture the students’ at-

titude what they do not recognize. It is preferable to

capture the students’ attitude by combining these two

approaches.

4.1 View-index for LS and GB

Firstly, we deﬁne the view-index of a student between

L and S (index for LS) is the average of the index

values of the terms she used in the answer texts (i.e.,

All). Formally, for s

∈ S (i = 1, 2, . . . , n) and for r ∈

{LS, GB}, we deﬁne v

) as follows:

) = mean

∈T

∑

∈T

∑

∈T

. (3)

By deﬁnition, v

) = ±1 iff ι(t

) = ±1 for all t

∈ T

respectively. The former case means that the student

uses the terms that appear only in L if r = LS, and

only in G if r = GB, and the latter case means she uses

only those terms in S if r = LS and those in B if r = GB.

Figure 3 shows the correlation between the view-

indexes of students for LS and GB. As is easy to see,

they are correlated strongly with the correlation co-

efﬁcient r = 0.78. Thus, we can say that the student

using the terms that are used mainly in evaluating lec-

tures tends to use the terms that are used more in good

evaluations than in bad evaluations. In other words,

the student using the terms that are relatively more

used for evaluating students themselves tends to use

the terms that are used rather in bad evaluations.

Even with such a high correlation between the

view-indexesfor LS and GB, some students have sim-

ilar view-indexes for LS and have quite different in

view-indexes for GB. For example, the view-indexes

for LS of St01 and St23 are similar; 0.71 for St01 and

0.75 for St23. For the view-indexes for GB, St01 has

0.77, which means that St01 seems to pay much at-

tention to good points. In comparison with St01, St23

has 0.29 in the view-index for GB, which is a little

bit greater than the average value of 0.137. Thus, we

may say that St23 evaluated from a more balanced

viewpoints than St01.

4.2 Investigation by Grouping

In order to clarify the differences between students,

we divide the students into groups, and compare them.

As we can see, the gap between a student and the next

one in their view-index for LS takes the maximum

value at the gap between St02 and St05. The view-

indexes of St02 and St05 are 0.44 and 0.31, respec-

tively. Thus, their difference, or gap, is 0.13. Now, we

have two groups LH and LL using the threshold value

0.4. The former group consists of the students who

have the LS view-indexes greater than 0.4, whereas

the latter consists of those who have less than 0.4.

As we recognize that each group seems to be con-

sisted with two subgroups divided by a relatively big

gap in each group. Therefore, we divide each group

into two subgroups at the gap having the biggest value

in the group. For group LH, the maximum gap is the

one between St20 and St18: the amount of the gap is

0.10. By using this gap, we divide the group LH into

two subgroups LHH and LHL. Similarly, we divide

A Characterization of Student’s Viewpoint to Learning and its Application to Learning Assistance Framework

625

ϮϬϭϳͬϯͬϭϯ

^ƚϬϮ

^ƚϬϱ

^ƚϭϴ

^ƚϮϬ

^ƚϭϮ

^ƚϮϭ

^ƚϬϯ

^ƚϭϬ

>,,

>>,

>,>

>>>

^ƚϬϭ

^ƚϮϯ

Figure 3: Correlation between view-indexes for LS (x-axis) and GB (y-axis).

Table 2: Statistical properties of the groups of students.

Prop.

LLL LLH LL LHL LHH LH

Size 3 19 22 2 5 7

LS view-index (×100)

Max

−0.4 31 31 47 75 75

Min −0.5 6 −0.5 44 57 44

Average

−2 21 18 46 67 61

Range

5 25 36 3 18 31

GB view-index (×100)

Max

5 39 39 55 77 77

Min

−13 -14 -13 17 29 17

Average −2 8 7 36 52 48

Range

18 53 40 38 47 59

Examination Score

Max

80 99 99 82 64 82

Min

75 29 29 66 27 27

Average 77.0 70.0 70.9 73.7 48.1 55.4

Range

5 70 70 16 37 55

the group LL into LLH and LLL by using the gap be-

tween St03 and St10. Note that we ignore the students

having the value 0 in both indexes for LS and GB be-

cause they did not answer at all to all the questions in

consideration.

Table 2 shows statistical information about the

groups. Regarding the range of the LS view-index,

the groups LL and LH have similar values. For sub-

groups, the ranges of the subgroups with upper value

are much larger than those with lower value in each

group of LL and LH.

Regarding GB view-index, LH group has much

higher values than LL group in terms of the maxi-

mum, minimum, and average values. Range for LH

is also bigger than that of LL group. Regarding sub-

groups, the upper subgroups LLH and LHH have big-

ger ranges than the lower subgroups LLL and LHL,

respectively. The ranges of upper subgroups are sim-

ilar.

Regarding examination scores, LL group has a lit-

tle bit higher values than LH group. As for subgroups,

the upper group has lower average examination score

and larger range in each groups LL and LH.

Considering some subgroups include just a small

amount of students, the results obtained in this analy-

sis may not be applicable to other data. However, the

analysis method itself should be applicable to other

data as well. From our experience in our previous

studies, characteristic feature(s) often become(s) clear

by dividing the members into groups, and compare

these groups.

Table 3 shows the 10 terms mostly used by the

student who have the maximum and the minimum

view-index values for GB in each subgroup. Note

that the terms with least frequency in each student

are part of the terms among those with the same fre-

quency. The terms in the table are in the form “En-

glish(Japanese):PoS”, where “Japanese” is the origi-

nal word in Japanese and “English” is its correspond-

ing English word or expression. The “PoS” part

shows the part of speech of the word. Note that

the English part is of the form “To+Verb” such as

“ToSearch” in “ToSearch():n” shows that the word is

“Sahen-noun”. Sahen-noun is a special type of noun

which turns into its verb form by adding “suru (mean-

ing do)” at the end of it. For example, by adding

“suru” to “ToSearch(, read Kensaku):n” we have the

verb from “Kensaku-suru (search-do)”, which means

“to search”.

Let us take St07, as an example, who belongs to

the LLL group, where the members used the mostly

A2E 2017 - Special Session on Analytics in Educational Environments

626

Table 3: Top 10 mostly used terms of the students with the maximum and minimum GB view-index values in each subgroup.

LLL LLH LHL LHH

Max St10 # St24 # St02 # St01 #

1 Time(Ł):adv 3 Do():v 8 ToSearch():n 4 Investigate():v 2

2 Assignment():n 3 Can():v 5 Lecture():n 3 Can():v 2

3 Setup():v 2 Library(Ł):n 5 Become():v 2 Know(m):v 2

4 Do():v 2 ToSearch():n 4 Information():n 2 Do():v 1

5 Do():v 1 Homework(h):n 4 Do():v 1 Some():n 1

6 ToExercise(K):n 1 ToIntroduce():n 4 Investigate():v 1 Proverb():n 1

7 Words(P):n 1 Not():o 3 Usually(i):adv 1 SoMuch():adv 1

8 Homework(h):n 1 NotMuch():adv 2 PC(PC):o 1 Amazing():adj 1

9 Study():n 1 Think(v):v 2 NotMuch():adv 1 Many():adv 1

10 0():v 1 ToDevise(Hv):n 2 0():adv 1 Beneﬁt():n 1

Min St07 # St12 # St18 # St23 #

1 Do():v 12 Not():o 4 Do():v 4 Become():v 1

2 Homework(h):n 7 Homework(h):n 3 Can():v 4 Opportunity(@):n 1

3 Think(v):v 6 Do():v 2 Become():v 3 No():adj 1

4 Lecturer():n 5 Become():v 2 ToSearch():n 3 Now():adv 1

5 Not():o 4 See():v 2 Homework(h):n 3 Homework(h):n 1

6 Exist():v 4 Think(v):v 2 ToIntroduce():n 3 See():v 1

7 Can():v 3 Time(Ł):adv 2 Not():o 3 Know(m):v 1

8 Me(ł):n 3 Many():adj 2 Lecture():n 2 Library(Ł):n 1

9 Contents(e):n 3 Can(o):v 2 Think(v):v 2 ToSolve():n 1

10 Good():adj 3 Not():o 2 Good():adj 2 ToPrepare():n 1

self-oriented, or self-referencing, terms. Her view-

index value for GB is the minimum in the group,

which means that the terms she used have tendencies

to be used in the texts for bad evaluation.

However, she does not disappointed with herself.

On the contrary, she seems to believe in herself. Her

answer text to (Q7), or SB, is as follows: “It was

too late to ﬁnd out the intention of the lecturer about

the homework assignments. Further, it is the worst

thing that I complained about it and what the lecturer

thought about. However, once I have recognized my

failure, I will not fail again. Truly it is my own re-

sponsibility what I can see and learn from what are

presented. Here after, I will infer the intention of oth-

ers and try to act accordingly. ” In the answer text,

she admitted what she did was wrong, and at the same

time, she decided to correct her way of action. Thus,

the use of terms for negative evaluation does not di-

rectly mean the student who uses is negative.

As the second example, let us choose St01, who

represents lecture-oriented and good-evaluation stu-

dents. The terms appearing in the list are generally

positive ones. Actually, she answered to the question

(Q2), or LG, only among the four questions. This is

the reason why she have such high values for view-

indexes for LS and GB.

As the third example, let us choose St10, who rep-

resents the most SG-oriented students, who are differ-

ent from the majority of students. Her answer to (Q6),

or SG, just praised herself that she did her homework

by taking a long time. The terms appearing in her

most frequently used terms list show she is highly in-

terested in the time and homework assignment.

As the fourth example, let us choose St23, who

represents lecture-oriented and bad-evaluation stu-

dents. She only answered to the lecture-related ques-

tions, i.e., (Q2) and (Q3). This is the reason why she

is lecture-oriented. Regarding the answer to (Q2), or

LG, her answer, and thus her viewpoint, is an ordinary

one. However, her answer to (Q3), or LB, is different

from other students. She mentioned a small quiz at

the beginning of every lecture, and complained about

it. There are no students who mentioned it, thus the

terms she used have strongly negative value in GB in-

A Characterization of Student’s Viewpoint to Learning and its Application to Learning Assistance Framework

627

^ƚϬϮ

^ƚϬϱ

^ƚϭϴ

^ƚϮϬ

^ƚϭϮ

^ƚϮϭ

^ƚϬϯ

^ƚϭϬ

>,,

>>,

>,>

>>>

^ƚϯϰ

^ƚϬϰ

^ƚϮϰ

^ƚϮϱ

^ƚϯϱ

^ƚϬϭ

^ƚϮϯ

^ƚϯϭ

^ƚϯϯ

^ƚϯϬ

^ƚϭϭ

^ƚϭϳ

^ƚϭϵ

^ƚϭϰ

^ƚϬϯ

^ƚϭϱ

^ƚϬϲ

^ƚϭϯ

^ƚϯϮ

^ƚϮϮ

^ƚϮϵ

^ƚϮϴ

^ƚϮϳ

^ƚϮϲ

^ƚϬϴ

^ƚϬϳ

^ƚϬϵ

Figure 4: Correlation between View-index for LS (x-axis) and Examination Score (y-axis) of Students.

dex, and here view-index value for GB becomes bad-

oriented accordingly.

4.3 Correlation Analysis between

View-index for LS and Examination

Score

Figure 4 shows the correlation between the view-

index for LS and examination score. As we can see,

they have a weak negative correlation with correla-

tion coefﬁcient r = −0.26. Thus, examination score

decreases as view-index for LS increases. In other

words, examination score increases as students care

more about themselves than lectures.

This result probably means that the students who

are more interested in their current status tend to retro-

spectively assess themselves, and thus, they do more

effort to correct their everyday attitudes. Such charac-

ters might be resulted in better academic performance

and increase of examination score.

Another notable ﬁnding in the ﬁgure is that the

range of examination scores are wide for those stu-

dents who have similar values in their view-index for

LS. For example, the maximum examination score in

LL group is 99 of St03 and the minimum examination

score is 29 of St32, and thus, the range is 70.

This result inspires that there are many factors that

relate to the examination score and student’s view is

one of them. The possible factors may include the

student’s potential ability in learning, amount of time

used for homework, the amount of concentration dur-

ing lectures, diligence in studying.

4.4 Data Analytics for AIS

In this paper, we have dealt with the questionnaire,

and investigated its answer texts. The questionnaire

was done when a series of lectures was almost ﬁn-

ished, and it asked the attending students to retrospec-

tively evaluate by looking back the lectures and the

students themselves.

In this section, we discuss how to improve the

questionnaire in order to make it more useful for LAF.

The eventual goal of our LAF is not only to provide

useful information to the lecturer and the supporting

staff so that they are able to advise students appropri-

ately, but also to help the students with understanding

their exact status, and the students themselves solve

their problems, and improve their academic perfor-

mance. However, we set the main role of LAF to an

information providing for advising students in learn-

ing at the moment.

The process from the questionnaire to the advice

to students consists of the following steps:

Step 1: Questionnaire → Answer-text

Step 2: Answer-text→ Viewpoint/Attitude to learning

Step 3: Viewpoint/Attitude → Advice for improve-

ment

In Step 1, the teacher of the class asks the stu-

dents to answer the questions. It is more convenient

to answer in a Web site than delivering question and

answer sheets. Regarding the frequencies of the ques-

tionnaire, we have asked once in a course at the end

of the lectures. In order to make LAS work more ef-

fectively, the questionnaire should be done more and

earlier. Our idea is to have questionnaire once in 3 to

4 lectures. Then, the teacher is able to obtain informa-

tion about students from the early lectures to the end,

so that he/she can give timely advices to the students.

A2E 2017 - Special Session on Analytics in Educational Environments

628

Step 2 is the analysis and analytics step. On the

basis of our previous studies and the study in Sec-

tion 4 are possible candidates for this step. We would

proceed our studies in this direction so that the results

are applied to the framework of LAF, and will be used

in this step.

In Step 3, the professors advice the students on the

basis of their viewpoint and attitude to learning, ob-

tained in Step 2. The advices will be provided to the

students as a whole, as well as to each student individ-

ually. Also, the understanding of students’ viewpoints

and attitudes will help the professors for improving

their teaching style.

5 CONCLUDING REMARKS

It becomes more and more important for universities

to motivate students in learning in order to increase

their academic performance. An aim of this paper

is to propose a learning assistance framework (LAF),

which consists of three types of human assistant: lec-

turer, supporting staff, data analyst, together with a

learning assistant system (LAS). Students are able to

study under the supports from LAF.

Another aim of this paper is to investigate how the

results of data analytics could be effectively used in

LAF. We showed some of our ﬁndings in our previous

studies and discussed howthey could be applied in the

services provided by LAF.

Further, as a part of developing methods for data

analytics, we pursued a case study of data analytics

for answer texts from contrasting questions. By using

the index deﬁned for measuring the balance of usage

between two contrasting texts, we have characterized

the students in their term usage.

The contributions of the study in this paper in-

clude not only the ﬁndings from speciﬁc data, but also

to show usefulness of the methods of analysis.

For further studies, we have to investigate the fol-

lowing topics: (1) To develop a method to devise new

ideas further, and to perform reﬁnement of dedication

to the study of student’s viewpoints, efforts, and atti-

tudes to learning. (2) By collecting data from a va-

riety of courses, to analyze them, and to verify if the

results of the study in this paper and those in our pre-

vious studies are also holds, or not. (3) To generalize

and formalize the analysis methods, and to integrate

them into an automated data analysis system, so that

at least a part of data analysis are performed automat-

ically and the data analyst are able to spend their time

in other topics.

ACKNOWLEDGMENT

This work was supported in part by JSPS KAKENHI

Grant Number JP15K00310.

REFERENCES

Ames, C. and Archer, J. (1988). Achievement goals in the

classroom: Students’ learning strategies and motiva-

tion processes. In Journal of Educational Psychology,

Vol.80, No.3, pp.260–267.

Blackboard. Anz blackboard. http://anz.blackboard.com/

[2017.2.10].

C. Romero, S. Ventura, P. E. and Hervas, C. (June 2008).

Data mining algorithms to classify students. In 1st In-

ternational Conference on Educational Data Mining

(EDM 2008), pp. 8–17.

Higuchi, K. Kh coder. http://khc.sourceforge.net/en/

[2017.3.13].

K. Goda, S. H. and Mine, T. (2013). Automated evaluation

of student comments on their learning behavior. In

12th International Conference on Advances in Web-

Based Learning (ICWL 2013), pp. 131–140.

Kudo, T. Mecab: Yet another part-of-speech

and morphological analyzer (in japanese).

http://taku910.github.io/mecab/ [2017.3.13].

Minami, T. and Ohura, Y. (2012a). An attempt on effort-

achievement analysis of lecture data for effective

teaching. In Database Theory and Application (DTA

2012), in T.-h. Kim et al. (Eds.): EL/DTA/UNESST

2012, CCIS 352, Springer-Verlag, pp. 50–57.

Minami, T. and Ohura, Y. (2012b). Towards development of

lecture data analysis method and its application to im-

provement of teaching. In 2nd International Confer-

ence on Applied and Theoretical Information Systems

Research (2ndATISR 2012), 14pp.

Minami, T. and Ohura, Y. (2013a). Investigation of stu-

dents’ attitudes to lectures with tex-analysis of ques-

tionnaires. In 4th International Conference on E-

Service and Knowledge Management (ESKM 2013),

7pp.

Minami, T. and Ohura, Y. (2013b). Lecture data analy-

sis towards to know how the students’ attitudes affect

to their evaluations. In 8th International Conference

on Information Technology and Applications (ICITA

2013), pp. 164–169.

Minami, T. and Ohura, Y. (2014). A correlation

analysis of student’s attitude and outcome of lec-

tures –investigation of keywords in class-evaluation

questionnaire–. In Advanced Science and Technology

Letters (ASTL), Vol.73 (FGCN 2014), pp. 11–16.

Minami, T. and Ohura, Y. (2015a). How student’s atti-

tude inﬂuences on learning achievement? –an analysis

of attitude-representing words appearing in looking-

back evaluation texts–. In International Journal of

Database Theory and Application (IJDTA), Science

& Engineering Research Support Society (SERSC),

Vol.8, No.2, 2015, pp. 129–144.

A Characterization of Student’s Viewpoint to Learning and its Application to Learning Assistance Framework

629

Minami, T. and Ohura, Y. (2015b). Towards improv-

ing students’ attitudes to lectures and getting higher

grades –with analyzing the usage of keywords in

class-evaluation questionnaire–. In The Seventh In-

ternational Conference on Information, Process, and

Knowledge Management (eKNOW 2015), pp. 78-83.

Moodle. Moodle - open-source learning platform. In

https://moodle.org/ [2017.3.13].

Moodle. Moodledocs. In

https://docs.moodle.org/dev/Main

Page [2017.3.13].

Romero, C. and Ventura, S. (2007). Educational data min-

ing: A survey from 1995 to 2005. In Expert Systems

with Applications, Vol. 33, Issue 1, pp. 135–146.

T. Minami, Y. O. and Baba, K. (2017). Estimating student’s

viewpoint to learning from lecture/self-evaluation

texts. In Proc. The Ninth International Conference on

Advanced Cognitive Technologies and Applications

(COGNITIVE 2017), Feb. 2017, pp. 38–43. IARIA.

A2E 2017 - Special Session on Analytics in Educational Environments

630