Research on Classroom Teaching Behavior Analysis System Based on

Artificial Intelligence Technology

Xin Zhou

College of Computer and Information Science, Chongqing Normal University, China

Keywords: Teaching Behavior Analysis System, Computer Vision, Natural Language Processing.

Abstract: So far, the mainstream of classroom teaching behavior analysis system is S-T analysis method, FIAS and

ITIAS, these three analysis systems are aimed at teaching behavior in teaching video. Through the objective

analysis of teaching behavior, we can make an accurate diagnosis of teaching practice, provide objective

guidance for teachers ' teaching, and ultimately provide an effective way for teachers ' professional

development. With the continuous development of artificial intelligence, computer vision and natural

language processing technology are gradually applied to the teaching behavior analysis system, which

solves the disadvantages of tedious data collection and large amount of calculation in the teaching behavior

analysis system. Based on three different classroom teaching behavior analysis systems, this paper

summarizes the current situation of teaching behavior analysis in the field of artificial intelligence, aiming

to provide reference for how to construct an efficient intelligent analysis system of classroom teaching

behavior based on artificial intelligence technology.

1 INTRODUCTION

During the 14th Five-Year Plan period, the Notice

on the Implementation of the National Training Plan

for Kindergarten Teachers in Primary and

Secondary Schools (Ministry of education 2020)

clearly stated that it is necessary to promote the

integration of artificial intelligence and teacher

training to help teachers develop with high quality.

In the face of ' one teacher, one excellent lesson '

this huge high-quality teaching resources, teachers

through observation can effectively improve the

teaching level, promote teachers to improve the

teaching process. However, this only stays on the

surface. Most observation teachers can only imitate

their gods but do not experience their shapes. The

teaching behavior analysis system can effectively

help the observation teachers to deeply understand

the interaction between teachers and students. In the

face of its disadvantages such as cumbersome data

acquisition, large amount of calculation, and high

labor consumption, Transana software and Nvivo

software have proposed that the efficiency and

accuracy are effectively improved through automatic

processing of a large number of codes (Zhang 2020).

With the maturity of computer vision and natural

language processing technology, the most

cumbersome data acquisition process can also be

automated. Artificial intelligence empowering

education (Jia 2018), facing three kinds of

classroom teaching behavior analysis system and

two kinds of artificial intelligence technology, how

to build efficient classroom teaching behavior

intelligent analysis system is attracting more and

more scholars ' attention.

2 CLASSROOM TEACHING

BEHAVIOR ANALYSIS

SYSTEM

2.1 S-T Analysis

With the continuous change of education, the

traditional classroom of single teaching mode has

been unable to carry the needs of modern teaching,

and the intervention of new technology has

promoted the emergence of smart classroom. With

the support of information technologies such as big

data and learning analysis, teachers can fully

implement diagnostic analysis and intelligent

resource push in the teaching process, and carry out

Zhou, X.

Research on Classroom Teaching Behavior Analysis System Based on Artiﬁcial Intelligence Technology.

DOI: 10.5220/0011910200003613

In Proceedings of the 2nd International Conference on New Media Development and Modernized Education (NMDME 2022) , pages 267-272

ISBN: 978-989-758-630-9

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

267

‘cloud + terminal’ learning activities and support

services. The smart classroom ecological

environment generally includes interactive screens,

freely assembled desks and chairs, tablets, cameras

and other hardware, as well as software such as e-

schoolbag systems and intelligent recording and

broadcasting systems (Li 2020). The intelligent

teaching equipment of smart classroom provides a

new situation for teachers' teaching and students'

learning. Teachers can simplify and optimize the

three teaching processes before, during and after

class according to their characteristics.

2.2 FIAS

FIAS was proposed by American scholar Ned

Flanders in the 1960s (Huang 2021). It believes that

language interaction is the main way of classroom

interaction, and 80 % of the teaching behavior in the

classroom belongs to teacher-student discourse

interaction. The system samples the classroom

records for 3 seconds according to its coding system,

and the data is recorded in the analysis matrix. By

analyzing the composition of the row and column

elements in the matrix, the classroom structure,

teacher tendency or style of the course can be

obtained.

2.3 ITIAS

With the continuous integration and development of

new technologies and education such as meta-

universe, artificial intelligence and big data, the

application of smart classrooms is becoming more

and more extensive. The ITIAS classroom

interaction system based on information technology

came into being on the basis of FIAS (Cen 2021).

On the one hand, ITIAS refines the category of

human interaction, on the other hand, it also expands

the interaction between human and technology, and

can effectively analyze the teaching behavior in new

education methods such as smart classroom.

2.4 Comparison of Three Teaching

Behavior Analysis Systems

Based on the analysis of the three teaching behavior

systems, their advantages and disadvantages are

summarized in Table 1.

Table 1 The advantages and disadvantages of three teaching behavior analysis systems

Advantages Disadvantages

S-T

analysis

Clearly defined teaching behavior reduces error

rate during data sampling

Focusing on teaching behavior, ignoring the

role of language in the classroom, and

teachin

behavior classification is too sim

FIAS

Rich dimensions of teaching behavior enable

observers to accurately identify and classify

classroom language behavior to a certain extent

Focus on linguistic analysis, ignoring

meaningful non-linguistic information

ITIAS

Focusing on the role of technology in the

teaching process, it is more suitable for new

educational methods such as smart classrooms.

There are too many categories of teaching

behaviors, which are prone to errors in

identification.

3 RESEARCH ON WAYS TO

ANALYSE TEACHING

BEHAVIOUR IN THE FIELD

OF ARTIFICIAL

INTELLIGENCE

Classroom teaching behaviour refers to the actions

taken by teachers and students in the classroom in

order to achieve certain teaching objectives (Yan

2017). Teaching behaviour can be divided into two

categories: action and speech. For these two

categories of teaching behaviour, computer vision

can be used to identify the action subject at the

current data collection point, and also natural

language processing technology can be used to

identify the discourse person at the current data

collection point. Therefore, teaching behaviour

analysis approaches under the AI domain are

divided into two categories based on computer

vision and based on natural language processing

techniques (Zhang 2021). At present, the theoretical

bases used for intelligent analysis systems of

teaching behaviour are all S-T analysis methods due

to their strong operational nature.

NMDME 2022 - The International Conference on New Media Development and Modernized Education

268

3.1 Computer Vision-Based Intelligent

Analysis System for Classroom

Teaching Behaviour

The main technique used to identify action subjects

in videos using computer vision is face detection.

Early face detection techniques were limited to a

single background and could only solve videos

where the face was facing the camera. With the

continuous improvement of technology, face

detection techniques now have three types of

methods based on knowledge, feature-based and

appearance-based (You 2017), and the most

commonly used method applied to classroom

teaching behaviour analysis is the appearance-based

AdaBoost method.

Zhou Pengxiao et al (Zhou 2018) obtained data

information from three aspects, such as the number

of faces, contour features and subject action

amplitude of the detected video frame images, used

Bayesian causal net model to reason about the

subject's behavioural features, obtained behavioural

sequences, constructed a teaching model for teaching

videos, and finally achieved the design of intelligent

recognition of S-T behaviour in classroom teaching

videos. Li Litao (Li 2017) converted the data

sampling points of the video into images, used

AdaBoost face detection technology and similarity

metric to obtain teacher behaviour, and attributed

data samples that were not teacher behaviour to

student behaviour according to the definition of S-T

analysis method, so as to obtain the S-T data

sequences in the corresponding teaching videos.

3.2 Analysis of Teaching Behaviour

Based on Natural Language

Processing

Mu Su (Mu 2019) proposed an S-T analysis method

based on speech similarity recognition from the

perspective of classroom speech, in which teachers

and students recorded vocal information in advance,

and voice clips were collected at fixed time intervals

to compare with the recorded vocal templates, so as

to determine the identity of the speaker of the

sample. This classroom discourse analysis system

based on vocal recognition technology can only get

accurate results after the vocal pattern model of

teacher and student discourse is entered in advance.

To address the above drawbacks, Guilin Liu (Liu

2020) established a classroom teaching corpus,

proposed a speaker classification algorithm for pre-

clustering recognition, and designed and developed

a teaching behaviour analysis system where the

input is a live classroom recording or audio based on

the S-T analysis method as the theoretical basis.

3.3 2x3 Model

Through the above analysis of the three teaching

behaviour analysis systems and two artificial

intelligence techniques, the 2X3 model (table 2) is

proposed in this paper. Where operability refers to

whether the development and design of the

intelligent analysis system can be achieved, and

practicality refers to whether the system, once

developed, can provide teachers with objective and

comprehensive guidance. As the S-T analysis

method of behavioural categorisation is too concise

to provide teachers with a referenceable meaning,

the practicality of an intelligent analysis system for

teaching behaviour, whether based on computer

vision or natural language processing techniques, is

poor. On the other hand, the ITIAS behavioural

categories are too complex, making computer

recognition difficult, and neither computer vision

nor natural language processing alone can define the

categories, requiring the integration of two artificial

intelligence techniques in the system, thus

increasing the burden of system design and

development. behaviour is derived from language,

for which natural language processing is somewhat

more tractable than computer vision.

Based on the 2X3 model (table 2), it can be

concluded that of the two artificial intelligence

technologies and the three teaching behaviour

analysis systems, the intelligent teaching behaviour

analysis system based on natural language

processing technology and FIAS is the most

efficient.

Table 2 2X3 model

Computer vision Natural language processing

S-T parsing Highly operational, poorly practical Highly operational, poorly practical

FIAS Weakly operational, highly practical Highly operational, highly practical

ITIAS Weakly operational, highly practical Weakly operational, highly practical

Research on Classroom Teaching Behavior Analysis System Based on Artiﬁcial Intelligence Technology

269

4 DESIGN OF AN INTELLIGENT

ANALYSIS SYSTEM FOR

TEACHING BEHAVIOUR

BASED ON NATURAL

LANGUAGE PROCESSING

AND FIAS

4.1 Theoretical Foundation

FIAS is known as Flanders Interaction Analysis

System (FIAS) in Chinese. It is based on its own

FIAS coding system, which is used to describe the

interaction behaviour between teachers and students

in the classroom. The recorded video is sampled

once every 3 seconds, the codes are recorded in a

matrix, and then the classroom behaviour is

analysed according to the meaning of the row and

column elements in the matrix. In this paper, we

take the 2018 ministerial-level excellent lesson

"Reflection of Light" in the One Teacher One

Excellent Lesson platform as an example and use

FIAS to do a detailed analysis of classroom teaching

behaviour.

4.1.1 Data Collection

Due to the subjective nature of the data collection

process, the three data collectors were required to

collect the data separately and then discuss and

determine the points of contention when

aggregating. A total of 874 coding sequences were

eventually collected, and this data was used to form

a Flanders migration matrix and a statistical table of

classroom interaction rates.

Figure 1 FIAS migration matrix

4.1.2 Analysis of Classroom Behaviour

(1) The teacher language ratio is the proportion of

teacher language to overall language, coded from 1

to 7. Therefore, the teacher language ratio is:

(85+18+52+114+234+56+1)/839≈0.667

(2) The student language ratio is the proportion

of the student language to the overall language,

coded as 8 and 9, so the student language ratio is:

(108+3)/ 839 ≈ 0.133

(3) The classroom silence rate is the proportion

of silence or confusion to overall language and is

coded as 10, so the classroom silence rate is:

168/839≈0.200

(4) Indirect and direct influences are both part of

the teacher's language, with expressions of emotion,

encouragement and praise, taking advice and asking

questions as indirect influences, and lectures,

instructions and criticism as direct influences. The

ratio of indirect to direct influence is therefore:

(85+18+52+114)/ (234+56+1) ≈ 0.924

(5) Positive reinforcement refers to the positive

guidance that teachers provide to students in the

classroom, such as expressing emotions,

encouraging praise and taking on board opinions,

which can motivate students in the learning process.

Negative reinforcement consists of both instruction

and criticism, and usually when there is too much

negative reinforcement, students become bored with

the class. The ratio of positive to negative

reinforcement is:

(85+18+52)/(56+1) ≈ 2.719

The basic structure of Reflection of Light can be

seen from the five data items above. Since the

course requires students to do hands-on experiments,

NMDME 2022 - The International Conference on New Media Development and Modernized Education

270

the teacher needs to explain the principles, steps and

precautions of the Reflection of Light experiment to

the students before they actually do it, so the

teacher's language accounts for the majority, about

66.7%. In the middle of the class, students were

discussing with each other or doing the experiment

quietly, and the class was silent or chaotic, so the

percentage of silence in the class was higher, about

20.0%, which is the normal range of the laboratory

course.

In the fourth data item, the ratio is less than 1,

which means that the indirect influence is less than

the direct influence, i.e. the teacher prefers to control

the classroom directly and carry out teaching in an

orderly manner according to the original teaching

design, without giving the classroom enough

flexibility.

In the fifth data item, positive reinforcement is

much higher than negative reinforcement, about

three times higher, indicating that teachers can

easily motivate students to actively participate in

teaching activities in the classroom and that such a

classroom is preferred by students.

4.2 Technical Foundation

FIAS focuses on speech analysis in the classroom

and therefore uses two key technologies: speech

recognition and word frequency analysis. From the

above data analysis, the classroom structure analysis

can be summarised into five main data segments:

teacher speech ratio, student speech ratio, classroom

silence ratio, indirect vs. direct influence ratio, and

positive vs. negative reinforcement ratio. With the

continuous development and optimisation of

artificial intelligence technology, these data can be

derived through speech recognition technology and

word frequency analysis, providing teachers with a

quick and efficient analysis of teaching behaviour in

the face of such a large and high-quality teaching

resource as One Teacher One Lesson, and helping

teachers to gain a deeper understanding of the

interaction between teachers and students in the

course.

4.2.1 Speech Recognition Technology

Among the physical properties of speech, it has four

elements: pitch, intensity, length and quality of

sound. Different speech therefore has different

spectra, and when speakers are different, computers

can produce and distinguish the vocal pattern of the

current speaker based on these four elements.

Nowadays, KDDI has already matured in the

application of speech recognition technology, which

has two sub-categories under voice recognition

technology and voice dictation technology. Using

voice recognition technology to distinguish between

teacher language and student language, three types

of values can be derived: teacher language ratio,

student language ratio and classroom silence ratio.

4.2.2 Word Frequency Analysis Method

Speech recognition and conversion of speech

information into text information can be

accomplished by calling the speech dictation API

interface provided by KDDI. Word frequency

analysis can be used to reveal the dynamics and

research progress of a discipline, and word

frequency analysis is an effective means of mining

text. The word cloud is developed using Nodejs and

JAVA, and the word segmentation tool uses the

Jieba natural language tool to perform segmentation

and word frequency analysis. By using a tool such

as UWI to count and analyse the number of

occurrences of important words in the teacher's

language and the students' language in the video, the

interaction behaviour between the teacher and the

students can be clarified, resulting in two types of

values: the ratio of indirect to direct influence, and

the ratio of positive to negative reinforcement.

5 CONCLUSION AND OUTLOOK

The aim of this paper is to propose a theoretical

study for the construction of an efficient intelligent

analysis system for classroom teaching behaviour

based on artificial intelligence technology. The 2X3

model is proposed through the analysis of three

teaching behaviour analysis systems and two

artificial intelligence technologies. From the 2X3

model, it can be concluded that the intelligent

analysis system of teaching behaviour based on

natural language processing technology and FIAS is

the most efficient. When it comes to actually

applying the system in practice, speech recognition

technology and voice recognition technology are

already mature and can be invoked directly.

Therefore, text analysis is the top priority, and how

to build a library of effective word pockets and

calculate two types of data, indirect to direct

influence ratio and positive to negative

reinforcement ratio, by the number of word

frequency occurrences will be the next research

direction.

Research on Classroom Teaching Behavior Analysis System Based on Artiﬁcial Intelligence Technology

271

REFERENCES

Fei-Fei Cen,Yu-Hui Ma,Lun-Yu Zhou. Research on

ITIAS-based interactive analysis system in high

school IT classroom [J]. Computer Knowledge and

Technology, 2021, 17(11):84-86.

Huang Miaomiao, Zhong Mei, Pan Xia, Yang Caixiang.

Research on teacher-student behavior in quality

classrooms based on FIAS [J]. Educational

Information Technology, 2021(03): 61-64.

Jia Jiyou. Artificial intelligence empowers education and

learning [J]. Journal of Distance Education, 2018,

(1):39-47.

Li Litao. Research and implementation of an intelligent S-

T teaching analysis system based on computer vision

[D]. Huazhong Normal University, 2017.

Liu Guilin. An intelligent teaching behavior analysis

system based on speaker classification algorithm [D].

Huazhong Normal University, 2020.

Mu Su, Hu Yongchun. A method for analyzing S-T

teaching process based on speech similarity

recognition:, CN105590632B [P]. 2019.

Sang Suling, Wu Renke, Zhao Pujie. A case study of

junior high school physics based on S-T analysis [J].

Journal of Ningxia Normal College, 2020, 41(01):109-

112.

The General Office of the Ministry of Education and the

General Office of the Ministry of Finance issued the

Notice on the Organization and Implementation of the

National Training Program for Primary and Secondary

School Kindergarten Teachers in 2020 [J]. Bulletin of

the Ministry of Education of the People's Republic of

China. 2020(03)

Yan Long. Classroom teaching behavior: connotations and

research framework[J]. Global Education

Perspectives, 2007, 36(S1):39-44.

You Qingqing, Chen Haiyun, Luo Jun, Wang Xiaoyi. A

review of face detection technology [J]. Wireless

Interconnection Technology, 2017(10):137-140.

Zhang Hui, Li Fengxia, Ma Xiufang. A comparative study

of the classroom interaction analysis tools of "Youcai"

[J]. China Education Informatization, 2020(02):79-82.

Zhang Zhigao, Li Jing. Research on the analysis of

teaching behavior in the context of artificial

intelligence [J]. Information Recording Materials,

2021, 22(02):214-215.

Zhou Pengxiao, Deng Wei, Guo Cultivation, Liu

Qingtang. Research on intelligent recognition of S-T

behavior in classroom teaching videos[J]. Modern

Educational Technology, 2018, 28(06):54-59.

NMDME 2022 - The International Conference on New Media Development and Modernized Education

272