Effects of Information Granularity on Health Education: An

Artificial Intelligence-Based Situational R-Map Analysis

Mikiko Oono

, Masaaki Ozaki

, Shreesh Babu Thassu Srinivasan

and Yoshifumi Nishida

Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology,

Aomi, Koto-ku, Japan

Department of Mechanical Engineering, School of Engineering, Tokyo Institute of Technology, Meguro-ku, Japan

Keywords: Information Granularity, Injury Prevention, Education, Artificial Intelligence.

Abstract: Unintentional injury is the leading cause of death among children in Japan and around the world. Enforcement,

engineering, and education⎯also known as the “three Es”⎯currently constitute the core approach to injury

prevention, and education plays a critical role in school safety. Providing information tailored to learners is

an essential factor allowing educators to provide effective education, and we believe that granularity is one

of the key factors for tailored messages. The purpose of this study is 1) to propose a situational R-Map analysis

method to manipulate the granularity of injury data and 2) to examine how granularity affects injury

prevention education design using this method. In the situational R-Map analysis method, the words contained

in each sentence of an injury situation description are transformed into 100-dimensional vectors using the

distributed representation method. A situation vector is created as the average of the word vectors in each

sentence. The dimension of the situation vector is reduced from 100 to 2 using the “t-SNE” method. Then,

we reordered these clusters in order of severity. To examine how granularity affects injury prevention

education design, we conducted a workshop to see whether information granularity affects the number of

preventive strategies devised by caregivers. We created a list of five bar- or slide-related injury situations

(coarse list) and a list of a list of 30 bar-related or 19 slide-related injury situations (fine list). All participants

first read the coarse list to devise and write down preventive strategies for each type of playground equipment.

Then, they read the fine list to see whether they had come up with any additional strategies after reading the

fine lists, and if so, to write them down. A total of 131 caregivers participated in the study and the results

suggest that the appropriate granularity depends on the type of equipment and the learner’s occupation and

can be evaluated using our proposed method.

1 INTRODUCTION

Unintentional injury is the leading cause of death

among children in Japan and around the world

(Statistics of Japan, 2022, Peden et al., 2009). The top

three leading causes of death in children aged 0–14

years are shown in Table 1. As such injuries are a

substantial burden to society, the Japanese

government has placed a high priority on reducing the

incidence of childhood injuries. A government

agency for children and family affairs was established

in 2023 to provide seamless support for children in

Japan, and injury prevention is one of the main

themes (Cabinet Secretariat, n.d.). In terms of injury,

https://orcid.org/0000-0002-3698-5329

https://orcid.org/0000-0003-3630-5562

school safety is an area that the Japanese government

has been working on in recent decades. According to

a report by the Japan Sport Council (JSC), in 2022,

nine child deaths were reported in elementary schools

and one in preschool, and approximately 72,000 and

282,000 injuries occurred in preschools and

elementary schools, respectively, with the medical

costs of each case exceeding 5000 JPY (ca. 40 USD

in 2023) (JSC, 2022).

Enforcement, engineering, and education⎯also

known as the “three Es”⎯currently constitute the

core approach to injury prevention. Because the

engineering approach typically requires no or

minimal individual actions (Peden et al., 2009), this

550

Oono, M., Ozaki, M., Srinivasan, S. and Nishida, Y.

Effects of Information Granularity on Health Education: An Artiﬁcial Intelligence-Based Situational R-Map Analysis.

DOI: 10.5220/0012724400003693

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Computer Supported Education (CSEDU 2024) - Volume 1, pages 550-556

ISBN: 978-989-758-697-2; ISSN: 2184-5026

approach is generally considered to be more effective

than enforcement and/or education. However, for

many types of injuries at schools, either no

engineering strategies are available, or they take a

long time to implement. Consequently, education

plays a critical role in school safety. With an increase

in awareness about its importance at a community

level, by January 2024, the International Safe

Community Certifying Center had certified 17 Safe

Communities (SCs) and 21 International Safe

Schools (ISSs) through the Safe Community and Safe

School Designation Program (Japan Institution for

Safe Communities, n.d.). The SC framework, which

was originally established in Sweden in 1970, seeks

to systematize community activities in order to

facilitate the creation of safe and secure communities

in conjunction with citizens (International Safe

Community Certifying Centre, n.d.). The ISS system

is a version of the SC applicable to schools.

Nowadays, injury prevention education directed at

schoolchildren is also expanding (Oono et al., 2018,

2022).

Table 1: The five leading causes of death among persons

aged 5–19 years in Japan, 2022.

Age

(years)

Ranking

1 2 3

Congenital

malformations,

deformities, and

chromosomal

abnormalities

(483)

Respiratory

disorders

specific to the

perinatal period

(202)

Unintentional

injury

(60)

1–4

Congenital

malformations,

deformities, and

chromosomal

abnormalities

(114)

Unintentional

injury

(59)

Malignant

neoplasm

(46)

5–9

Malignant

neoplasm

(89)

Congenital

malformations,

deformities, and

chromosomal

abnormalities

(29)

Unintentional

injury

(28)

10–14

Suicide

(119)

Malignant

neoplasm

(84)

Unintentional

injury

(34)

* Numbers in parentheses indicate the number of cases.

One of the most difficult, but critical, factors for

injury prevention through education is to convince

people of the importance of injury prevention. This is

closely related to the concept of perceived

susceptibility in the field of health education.

Perceived susceptibility is a construct of the Health

Belief Model (HBM) developed by Irwin M.

Rosenstock (Glanz et al., 2008). The HBM is

concerned with the likelihood of contracting a disease

or developing a condition. According to the HBM,

people are more likely to practice healthy behaviors

if they believe that their chance of developing a

negative health condition is high (Glanz et al., 2008).

By applying this concept to injuries, people are most

likely to adopt preventive actions only if they are

convinced that serious injury can occur. Educating

caregivers, including schoolteachers, parents, and

staff in after-school programs is more complex, as

their perception of injury is ameliorated by the fact

that the injury does not affect them directly, but rather

the children in their care.

We have been conducting injury prevention

research for more than 20 years, and it is now well

understood from our experience that people’s

perceived susceptibility to injury depends on their

level of childcare experiences. For instance, when we

discuss an injury related to a swing, people with many

care experiences can think of a much wider variety of

injury risks than can novices. Therefore, from the

perspective of health educators, we need to provide

more details to novice caregivers regarding how

injuries occur so that they can assume and prepare for

prevention adequately. Our underlying assumption in

the present study is that information granularity is

critical to maximizing one’s learning ability, which is

the hypothesis tested in this study. Transmitting

information tailored to learners is always an essential

factor for educators who wish to provide effective

education (Hawkins, 2008), and we believe that

granularity is one of the key factors for such tailored

messages. In the present paper, we first propose a

situational R-Map analysis method to manipulate the

granularity of injury data. Then, we examine how

granularity affects injury prevention education design

using this method. Lastly, we introduce the Egao

search system using the proposed method.

2 AN ARTIFICIAL

INTELLEGENCE (AI)-BASED

SITUATIONAL R-MAP

We developed a situational R-Map analysis by

integrating existing R-Map analysis methods and text

mining. Conventionally, a R-Map is a well-known

method for prioritizing injuries from the viewpoints

of both frequency and severity (Ministry of Economy,

Trade and Industry, 2011). In this method, the words

contained in each sentence of an injury situation

description are transformed into 100-dimensional

vectors using the distributed representation method

Effects of Information Granularity on Health Education: An Artiﬁcial Intelligence-Based Situational R-Map Analysis

551

(word2vec). A situation vector is created as the

average of the word vectors in each sentence. The

dimension of the situation vector is reduced from 100

to 2 using the “t-SNE” method (Maaten and Hinton,

2008).

We used injury data from the JSC; the JSC

database is the largest school injury dataset in Japan.

Figure 1 shows a t-SNE graph depicting a cluster

analysis (20 clusters) of 6549 bar-related injury cases

that occurred in elementary schools using a k-means

approach. Each dot indicates one injury situation, and

each color indicates a similar injury situation.

Figure 1: A t-SNE graph depicting a cluster analysis of

6549 bar-related injury cases.

Next, we reordered these clusters in order of severity.

In the present study, we defined high-risk injuries as

those having both high frequency and high medical

costs. Figure 2 shows a situational R-Map for bar-

related injury cases based on the cluster analysis.

Figure 2: Situational R-map for bar-related injury cases

based on the cluster analysis.

Based on each cluster, we identified typical injury

situations. When situations from different clusters

were similar, we considered them to be the same

situation. Table 2 shows the top five examples of the

cluster situations based on injury severity. Using an

AI-based situational R-Map analysis makes it

possible to understand typical injury situations in a

data-driven way.

Table 2: Top five examples of cluster situations.

Cluster

No.

Injury situation

1 5, 9 Landed on one’s hands after falling.

2 1, 10 Hit one’s back and/or face after falling.

3 19 Hit one’s face on an unnoticed bar while

playing tag.

4 12 Failed to land properly and twisted one’s leg.

5 15 Some sand blew into one’s eyes or someone’s

foot touched one’s eyes.

3 EVALUATION OF HOW

INFORMATION

GRANULARITY AFFECTS

INJURY PREVENTION

EDUCATION DESIGN

In terms of injury prevention education, on the one

hand, vague information does not help much when

thinking about preventive measures. On the other

hand, too many details are ineffective because

learners cannot process them to develop appropriate

prevention strategies. The aim of this evaluation

study is to clarify how information granularity affects

injury prevention education design.

3.1 Method

3.1.1 Controlling Information Granularity

In this study, we controlled information granularity

by changing the number of clusters based on our

proposed method. Figure 3 shows a t-SNE graph of a

cluster analysis for bar-related injuries of 4480 cases

when determining the numbers of clusters as five

(left) and 30 (right). Then, we identified typical injury

situations for each cluster and reordered them in

terms of severity, which yielded two lists (Tables 3

and 4). Interim listings were omitted in the case of the

30 clusters in Table 4. As shown in these two tables,

identified typical injury situations are less detailed

when the number of clusters is five. We created two

lists for bar- and slide-related injuries. In the case of

slide-related injuries, we determined that the number

of clusters was 20 using a k-means approach.

CSEDU 2024 - 16th International Conference on Computer Supported Education

552

Figure 3: A t-SNE graph of a cluster analysis of bar-related

injuries (No. of clusters: 5 and 30).

Table 3: Injury situation list (No. of clusters: 5).

Injury situation

1 Fell on one’s hand or elbow and broke bones.

2 Fell and hit one’s body.

3 A child felt pain and a teacher noticed an injury.

4 Crashed into a bar when running nearby.

5 Fell and hit one’s face or a part of one’s body on bars or

the ground.

Table 4: Injury situation list (No. of clusters: 30).

Injury situation

1 Slipped and fell while moving forward and backward on

the bars.

2 Lost balance and fell on the arms or hands while

attempting to sit on the bars.

3 Fell to the ground.

4 Fell when unsupervised by a caregiver.

5 Hands slipped off the bars; failed to grasp the bars and

fell to the ground.

6 Fell and hit one’s arms. Complained about the pain.

7 Complained about the pain after a short time.

8 Failed to land properly and hit one’s body on the

ground.

9 Hit one’s face on a bar when playing tag.

10 Hit one’s face on a bar while moving through or failing

to notice.

28 Hit one’s mouth on a bar and started bleeding from the

teeth and lips.

29 Hit one’s head on a bar while jumping around.

30 Some sand blew into one’s eyes.

3.1.2 Analysis of the Influence of

Granularity on Injury Prevention

Education Design

In this subsection, we describe how we quantitatively

analyzed the influence of information granularity on

injury prevention education design. Where the

probability that event 𝑥

, 𝑥

…., 𝑥

𝑛

can be (𝑥

), (𝑥

),….

𝑃(𝑥

𝑛

), the average information H can be calculated as

Eq. (1). In this study, we calculated the average

information H for situations in clusters and used it as

an indicator to determine the granularity of injury

situations quantitatively. This average information

for injury situations (SH) is defined as Eq. (2).

)

Thus, the larger the average information for injury

situations (SH), the finer the granularity of the injury

situation information categorized into clusters. Table

5 shows a summary of the average information for

injury situations.

Table 5: Summary of the average information for injury

situations.

SH No. of clusters

2.30 5

4.84 30

Slide 2.29 5

Slide 4.17 19

3.2 Data Collection

We conducted a workshop to examine whether

information granularity affects the number of

preventive strategies devised by caregivers. The

workshop procedures were as follows. First, all

workshop participants attended a lecture on the

importance of injury prevention. Second, all

participants received and read a list of five bar- or

slide-related injury situations (coarse list). After

reading, we asked the participants to devise and write

down preventive strategies for each type of

playground equipment to the best of their ability.

Third, we gave the participants a list of 30 bar-related

or 19 slide-related injury situations (fine list). Then,

we asked whether they had come up with any

additional strategies after reading the fine lists, and if

so, to write them down. We conducted three

workshops: one for bar-related and two for slide-

related injuries.

3.3 Results

A total of 131 caregivers participated in this study. A

summary of the three workshops is shown in Table 6.

When we counted the number of preventive strategies

that participants had listed by reading the coarse lists,

on average, group 1 listed 2.7 strategies (range, 0–7),

group 2 listed 3.81 (range, 0–8), and group 3 listed

3.44 (range, 1–6). When asked to come up with

Effects of Information Granularity on Health Education: An Artiﬁcial Intelligence-Based Situational R-Map Analysis

553

additional strategies by reading the fine lists, on

average, group 1 listed 1.41 new strategies, group 2

listed 3.30, and group 3 listed 2.07. Overall, 82% of

the participants successfully came up with new

prevention strategies after reading the fine lists.

Table 6: Summary of the three workshops.

Group

No.

Occupation No. of

participants

Equipment

type

Mode

1 Daycare

teache

65 Bar Online

2 Children’s

center

teache

37 Slide On site

3 Daycare

teache

29 Slide On site

Figure 4 shows the relationship between the

average information for injury situations (SH) and the

number of strategies devised by the participants.

Regarding the number of strategies after reading the

fine lists, we added the numbers from the coarse and

fine lists. This figure clearly shows differences in

granularity effects by equipment type. As shown in

Figure 4, the slope connecting the two points for the

bars regarding fine and coarse injury situations is 0.56,

whereas the slope for the slide is 1.10. This means

that the participants devised more preventive

strategies for the slide after reading the fine lists.

Figure 5 shows that the slope connecting the two

points for the slide regarding fine and coarse injury

situations is 1.10 for daycare teachers, compared with

1.75 for children’s center teachers. This means that

the fine lists were more effective for allowing

children’s center teachers to think about preventive

strategies for slide safety.

Figure 4: Relationship between the average information for

injury situations (SH) and the number of strategies

(equipment type difference).

Figure 5: Relationship between the average information for

injury situations (SH) and the number of strategies

(occupation difference).

3.4 Discussion

In this study, we controlled information granularity

by changing the number of clusters based on our

proposed method and examined how it affected the

development of preventive strategies. The number of

strategies devised by learners is a critical factor in

injury prevention education design.

As shown in Figure 4, the fine lists for the bars

were less effective than the coarse lists in terms of

thinking about injury prevention strategies. This

might be because most bar-related injury situations

are nearly equal to “falling off the bars” or “hitting

the bars”. These situations cannot be more detailed,

even if the AI divides them into different clusters. By

contrast, slide-related injury situations vary, and thus

work well for learners to think about their actions.

Our results also suggest that granularity is affected by

the learner’s occupation. In this study, the fine lists

were more effective for children’s center teachers

than for daycare teachers. There are two important

points that should be taken from these findings. First,

our proposed method appears to be useful for

quantitatively evaluating the effectiveness of

educational information for learners. Health

educators often discuss the importance of tailored

information, and our method appears to be useful to

achieve these goals. Second, the appropriate amount

of granularity appears to depend on the type of

equipment and the learner’s occupation. This could

also help health educators create information tailored

to learners.

This study had several limitations. First, we did

not ask the participants about their age, sex, or

working experience, which could have affected the

CSEDU 2024 - 16th International Conference on Computer Supported Education

554

number of preventive strategies they devised. Second,

we only asked the participants about two types of

equipment; the findings may have differed if we had

asked about other types. Third, we did not randomly

assign the participants to evaluate the granularity

effects. Despite these limitations, the results of the

present study suggest that information granularity

affects the design of injury prevention education, and

that our proposed method is useful to assess such

granularity from the viewpoints of equipment type

and learner’s occupation.

4 DEVELOPMENT OF THE

EGAO SEARCH SYSTEM

“Egao” means smile in Japanese. We named our

system Egao because a smile is one of the most

important social assets of children. The Egao search

system (hereafter, the Egao system) has three

functions, as shown in Figure 6: 1) it facilitates user

searches for high-risk injuries, 2) it provides safety

tips, and 3) it allows users to ask questions of experts

and other Egao users. A login ID and password are

necessary to access the Egao system. The first

function is implemented by a situational R-Map

analysis. Egao users can search for injuries based on

the locations at which the injuries are known to occur,

such as classrooms, gyms, or hallways, the types of

playground equipment that are known to cause the

types of injury, or the types of sporting activities

known to cause such injuries. Options can be selected

by clicking on an icon. The users also can input text

describing the cause of the injury to search for

associated injury types. The current list of icons is

shown in Figure 7. By clicking the search button, a

list of injury cases is displayed in descending order of

risk. The implementation of this function is as

follows. First, by using word2vec, a text mining

technique, sentences in an injury database that

contain descriptions of how various injuries occurred

are transformed into numerical vectors. We used

injury data from the JSC for the last 5 years

(approximately 990,000 injury data points). Second,

the Egao system then transforms the sentences

inputted by Egao users into numerical vectors, as in

the first step. Third, the Egao system calculates the

cosine similarities between the two vectors, using the

Approximate Nearest Neighbors Oh Yeah (Annoy)

algorithm (Spotify, 2022), and then extracts the top

20 most similar injury situations. Fourth, the Egao

system reorders the 20 cases in order of severity.

Figure 6: Screenshot of the Egao search system menu.

Figure 7: Screenshot of the Egao search system.

The top five examples of extracted injury cases with

information on the nature and severity of the injury

are as follows. These representative examples show

the results in the Egao system for the sentence, “I was

going up the stairs, and I slipped and fell because the

floor was wet”.

● After cleaning the classroom, the student was

going up the stairs, slipped, and fell. The student

hit their hand on the floor and was injured.

(Broken bone, Severity: high)

● During break time, the student hurried down the

stairs. The student slipped and fell because the

floor was wet and hit their head and right knee

on the floor. The student was pale and in shock

for a while. (Bruise, Severity: high)

● The student was going down the stairs on the

way to the gymnasium to play with friends

during a lunch break. The student did not notice

that the floor was wet, and slipped and fell. The

student hit their head on the handrail and fell

down several steps until reaching the landing.

(Bruise, Severity: high)

● The student slipped and fell off the first step at

the bottom of the stairs after descending the

Effects of Information Granularity on Health Education: An Artiﬁcial Intelligence-Based Situational R-Map Analysis

555

stairs while holding their friend’s hand. (Broken

bone, Severity: high)

● During a short break between classes, the

student tripped and fell and hit their face on the

floor with considerable force. (Laceration,

Severity: high)

Here, we describe only five examples of injuries, but

as mentioned previously, users of the Egao system

obtain 20 examples of high-risk injury cases based on

their search terms.

In the second function of searching for safety tips,

Egao users can search for safety tips based on the

locations where the types of injury are known to

occur, such as in classrooms, gyms, or hallways, the

types of playground equipment on which they occur,

or the types of sports activities that cause them. These

options can be selected by clicking on an icon. Users

also can input words or sentences describing the

cause of injury to search for related safety tips. When

the Egao system cannot find safety tips in the list

based on the user’s search terms, it prompts the user

to ask questions; this function is explained as follows.

In the user questions function, Egao users can ask

questions regarding preventive actions. They can also

upload pictures and explain why they consider the

situations depicted in the images to be dangerous.

After posting the questions, they can receive

suggestions and comments from injury prevention

experts or other Egao users.

5 CONCLUSION

In this study, we conducted a situational R-Map

analysis by integrating R-Map analysis and text

mining methods. We applied the concept of

information granularity to injury prevention

education design to evaluate messages tailored to

learners. We controlled the granularity of injury

situation texts for this particular study, but this

strategy can be used for any kind of health

information to create effective health messages. We

also introduced the Egao system, which can be used

by schoolteachers. We plan to implement this system

in schools so that students and teachers can have fun

together learning about injury prevention. Most

importantly, by revealing that information granularity

influences learners’ ability to think about preventive

strategies, the results of this study indicate that

information granularity affects injury prevention

education design.

ACKNOWLEDGMENTS

This work was supported by the JST-Mirai Program

Grant No. JPMJMI22H3.

REFERENCES

Oono, Mikiko & Nishida, Yoshifumi & Kitamura, Koji &

Kanezaki, Asako & Yamanaka, Tatsuhiro. (2018).

“Change the Changeable” Framework for

Implementation Research in Health. 361-368.

10.5220/0006691303610368.

Oono M, Nishida Y, Kitamura K, Yamanaka T. Injury

Prevention Education for Changing a School

Environment Using Photovoice. Health Promot Pract.

2022 Mar;23(2):296-304. doi:

10.1177/15248399211054772. PMID: 35285323.

Cabinet Secretariat. (n.d.) Kodomo seisaku no suishinn

[Development of government strategies for children].

https://www.cas.go.jp/jp/seisaku/kodomo_seisaku_sui

shin/index.html

Glanz, K., Rimer, B. K., & Viswanath, K. (2008). Health

Behavior and Health Education: Theory, Research, and

Practice (4th ed.). Jossey-Bass.

Robert P. Hawkins, Matthew Kreuter, Kenneth Resnicow,

Martin Fishbein, Arie Dijkstra, Understanding tailoring

in communicating about health, Health Education

Research, Volume 23, Issue 3, June 2008, Pages 454–

466, https://doi.org/10.1093/her/cyn004

International Safe Community Certifying Centre. (n.d.).

https://isccc.global/

Japan Institution for Safe Communities. (n.d.). Nihon no

Sefu Sukuru [Safe School Japan]. https://www.jisc-

ascsc.jp/ss_japan.html

Japan Sport Council. JSC (2002). Gakkou no kannrika no

saigai [reiwa3nenndo ban] [Disaster under school

environment. The Version of 2021].

https://www.jpnsport.go.jp/anzen/kankobutuichiran/ka

nkobutuichiran/tabid/1988/Default.aspx

L.V.D. Maaten and G. Hinton (2008) Visualizing Data

using t-SNE. Journal of Machine Learning Research

9:2579-2605.

Ministry of Economy, Trade and Industry. (2011). Risuku

asesumento handobukku [Risk assessment handbook].

https://www.meti.go.jp/product_safety/recall/risk_asse

ssment_practice.pdf

Peden, M., Oyebite, K., Ozanne-Smith, J., Hyder, A. A.,

Branche, C., Rahman, F. A., Rivara, F., & Bartolomeos,

K. (2009). World Report on Child Injury Prevention

(1st ed.). World Health Organization.

spotify. (n.d.). GitHub - spotify/annoy: Approximate

Nearest Neighbors in C++/Python optimized for

memory usage and loading/saving to disk. GitHub.

https://github.com/spotify/annoy

Statistics of Japan. (2022). Portal Site of Official Statistics

of Japan. https://www.e-stat.go.jp/en/stat-

search/files?page=1

CSEDU 2024 - 16th International Conference on Computer Supported Education

556