Research on the Innovation of English Teaching Mode in Colleges and Universities under the Background of Digital Transformation

The application of information technology and digital technology has reshaped the teaching ecosystem and the relationship between various players, and the teaching rules, strategies and means need to be re-examined [1-3]. Under the background of digital transformation, the traditional college English teaching mode faces many challenges. The traditional English teaching mode in colleges and universities mainly relies on teachers’ lectures and students’ classroom participation, and lacks flexible and personalized teaching methods [4-6]. The digital transformation of education has brought new development opportunities for English teaching in colleges and universities. The wide application of digital teaching platforms, online learning resources and intelligent teaching tools provides strong support for the change and innovation of English teaching in colleges and universities [7-9]. The digital teaching platform provides a convenient teaching environment for teachers and students, which can realize the sharing and optimization of teaching resources, make the college English listening, speaking, reading and writing courses more vivid and practical, so as to improve the learning effect and interest of students, and through the online classroom, discussion forums, homework submissions and other functions, it can realize the interaction and communication between students and teachers, and help students to comprehensively improve their English language skills [10-11]. In the era of digital economy, personalization has become an important demand and trend, and English teaching in colleges and universities is no exception. The digital transformation of education provides more possibilities for personalized teaching. By analyzing students’ learning data and behaviors, teachers, with the help of intelligent teaching tools, can provide personalized learning content and adopt personalized teaching methods according to students’ different needs and levels, so that every student can get a better education [12-13]. In addition, in the context of digital transformation of education, assessment and feedback also show a new trend, the traditional examination and assessment methods have been difficult to adapt to the needs of teaching in the digital era, the digital teaching platform can collect students’ learning data and performance, through data analysis and intelligent assessment tools, teachers can obtain more accurate and comprehensive assessment results, and students can also get timely feedback and guidance, better self-adjustment and improvement [14-15].

Literature [16] aims to explore the constraints to innovation attempts by university English teachers located in Southwest China. Key determinants that emerged included time pressure and competing priorities. Skepticism about new ideas; lack of resource investment; primacy of textbooks; students’ language proficiency. And opportunities for professional development. Literature [17] uses PBL teaching mode for teaching, and the results show that the effect of PBL teaching mode is significantly better than the traditional teaching mode, which is conducive to improving students’ independent learning, inquiry learning, the ability to obtain, process and use statistical information, as well as the comprehensive ability to analyze and solve problems. Literature [18] selects youtube videos that show a variety of special spoken discourses and interactions to expand the sociolinguistic experience of Saudi English language students. It was found that: Learning spoken English using YouTube played a key role in improving the proficiency of Saudi students in English. The academic performance of the students in the experimental group was positively improved by the video instruction. Literature [19] lists some of the most interesting pedagogical challenges faced by ESP practitioners and suggests possible solutions, which are partly derived from overviews of relevant academic literature and partly from successful practical implementations. Literature [20] suggests that vr-based wearable devices create a highly immersive learning environment where self-confidence and relevance are positively correlated with academic performance, and are more likely to help learners find the right words, improve their accuracy, and increase their confidence and sense of achievement throughout the learning process than traditional teaching methods. Literature [21] focuses on a blended learning model represented by the online MyEnglishLab, which complements the English paper-based textbook. A complete study is presented, in which an undergraduate English course at the Faculty of Science of the University of Kralove, Hradec Králové, was conducted, applying the above mentioned pedagogical design and validating the effectiveness of the MyEnglishLab tool.

The article first researches the basic theory of knowledge graphs. Then a complete course knowledge graph is constructed based on the Neo4j graph database to store the knowledge, and the named entity recognition model of RoBERTa-wwm-BiGRU-GlobalPointer is used to combine the local and global information in this article. Further based on Attention+BiLSTMj for relationship extraction, a multimodal entity linking method is proposed to link with the matching of knowledge point entities through text matching, and finally a multimodal retrieval platform for courses based on knowledge graph is built. Finally, the English teaching classroom of a university is used as an empirical test object to analyze the satisfaction survey of the teaching experiment effect in depth, in order to verify the feasibility of the application of Knowledge Graph based on the English teaching classroom of the university, so as to further improve the design of the English teaching mode of the university.

2

Method

The application of knowledge mapping in the era of “Intelligence +” provides a new path for English curriculum and teaching design, which can conceptualize and structure English knowledge, help learners analyze and understand the logical relationship between knowledge points with the help of visualization tools, and promote the innovation of the English teaching mode in colleges and universities while enhancing the learning experience and effectiveness of learners. The effectiveness of visualization can be improved.

2.1

Knowledge mapping

2.1.1

Basic Theory

A knowledge graph is a structured representation of facts consisting of entities, relationships, and semantic descriptions.The knowledge graph is a network of knowledge bases that are connected through relationships between entities that have attributes. From a graph perspective, a knowledge graph is essentially a conceptual network in which nodes represent entities or concepts in the physical world, and various semantic relationships between entities form edges in the network. Specifically, a knowledge graph is a graph of directed attributes labeled. In the knowledge graph, each node has several attributes and attribute values, the edges between entities represent the relationships between nodes, the direction pointed by the edges represents the direction of the relationship, and the markers on the edges represent the type of relationship. The knowledge graph’s basic structure is depicted in Figure 1 below.

2.1.2

D3.js Framework

D3 is a JavaScript library that uses Web standards for data visualization centered around using drawing instructions to transform data, create new graphical data from source data, generate SVG paths, and create data visualization elements from data and methods in the DOM. Because it’s essentially JavaScript, you can do just about anything with JavaScript, especially in the area of data visualization, and using D3 can reduce the complex steps of generating a visualization to a few simple functions that can be converted into all sorts of gorgeous graphics with just the input of some simple data. For Echats, D3 is much more flexible than highly encapsulated charting libraries such as Echarts, and much simpler than low-level plotting libraries such as Processing. That is to say, D3.js provides users with a kind of library that is neither a special bottom layer nor a high-level package, making it much easier to use.

2.2

Entity Recognition Model

2.2.1

The RoBERTa-wwm-BiGRU-GlobalPointer model

The RoBERTa-wwm-BiGRU-GlobalPointer model is used in this section to accomplish the task of interrogative entity recognition. By introducing RoBERTa-wwm to pre-train the language model to obtain semantic feature vectors and improve the language comprehension ability of the model, BiGRU is used to bi-directionally encode the semantic feature vectors in order to obtain the complete contextual information and improve the prediction speed, and then GlobalPointer is used to solve the problem of entity nesting and improve the correctness of the label sequence. The structure of the RBGGP interrogative entity recognition model constructed in this paper is shown in Figure 2.

2.2.2

Word Embedding Layer

The RBGGP model uses RoBERTa-wwm, which is mentioned in Chapter 3, as a pre-trained language model.By training the improved language model for Chinese, better-quality word vectors were obtained.The overall semantic comprehension of the model has also improved.The computational procedure is shown in Equation (1) and Equation (2), and the significance of the parameters is not repeated: (1) $h_{0} = W_{s} + W_{p}$ \[{{h}_{0}}={{W}_{s}}+{{W}_{p}}\] (2) $h_{α} = T r a n s (h_{α - 1}), α \in [1, N]$ \[{{h}_{\alpha }}=Trans\left( {{h}_{\alpha -1}} \right),\alpha \in \left[ 1,N \right]\]

2.2.3

Feature extraction layer

Although RoBERTa-wwm already has access to relatively rich semantic information, the contextual information of the context also plays a very important role in the prediction of entities in the named entity recognition task. Since the model will be used to parse user interrogatives, the faster the model predicts, the better.

The Gate Recurrent Unit GRU has a similar structure and performance to LSTM, but has a faster computational speed, the structure of GRU is shown in Figure 3.

The difference between GRU and LSTM is that it reduces the number of “gates” and relies only on hidden state memory and transfer of information, and has a simpler computational process than LSTM, which is shown below: (3) $z_{t} = σ (W_{z} \cdot [h_{t - 1}, x_{t}])$ \[{{z}_{t}}=\sigma \left( {{W}_{z}}\cdot \left[ {{h}_{t-1}},{{x}_{t}} \right] \right)\] (4) $r_{t} = σ (W_{r} \cdot [h_{t - 1}, x_{t}])$ \[{{r}_{t}}=\sigma \left( {{W}_{r}}\cdot \left[ {{h}_{t-1}},{{x}_{t}} \right] \right)\] (5) $s_{t} = \tanh (W_{s} \cdot [r_{t} \times h_{t - 1}, x_{t}])$ \[{{s}_{t}}=\tanh \left( {{W}_{s}}\cdot \left[ {{r}_{t}}\times {{h}_{t-1}},{{x}_{t}} \right] \right)\] (6) $h_{t} = (1 - z_{t}) \times h_{t - 1} + z_{t} \times s_{t}$ \[{{h}_{t}}=\left( 1-{{z}_{t}} \right)\times {{h}_{t-1}}+{{z}_{t}}\times {{s}_{t}}\] where z_t represents the update gate, which corresponds to the input and forgetting gates of the LSTM and shares the same function with them to update the hidden state of the whole network. r_r is the reset gate, which is used to selectively forget the previous time’s state h_t–1 the part that is not important for the current time’s state. x_i represents the input at the current time, s_t represents the state information at the current time, W_z, W_r and W_r represent the weights.

This model will use BiGRU to re-extract features from word embedding vector h_s and splice it with the original h_s to obtain the final feature H representing the whole text, which provides rich semantic support for the global pointer of the annotation layer. The computational process is shown in Eq. (7) and Eq. (8): (7) $l = B i G R U (h_{N})$ \[l=BiGRU\left( {{h}_{N}} \right)\] (8) $H = [h_{N}, l]$ \[H=\left[ {{h}_{N}},l \right]\] where h_w represents the word embedding vector output from the RoBERTa-wwm model; BiGRU () represents a series of computations in the BiGRU network; l represents the output vector of the BiGRU network; and H represents the final result after stitching h_N and l.

2.2.4

Entity labeling layer

Through the BiGRU layer, the contextual context information under long distance has been able to be obtained, but since BiGRU does not pay attention to the correctness of the labeling strategy, if the sequence is labeled only on the basis of the output of the BiGRU layer, it is very likely that there will be some prediction results that do not conform to the labeling strategy. The predictive labeling of text sequences generally adopts CRF, which can consider the probability of labeling from a global perspective and can ensure the correctness of the final results. However, the efficiency of CRF is relatively low and it cannot solve the problem of entity nesting well, this model will choose the Global Pointer Global Pointer, which is mentioned in Chapter 3, to accomplish the labeling task.

In the extraction task, there is a total of e entity type, then in the annotation layer, there are e layers of global pointers, and the multi-head processing is used to let one layer of global pointers to deal with a separate one entity type, so that the problem of entity overlapping can be effectively dealt with.

If the text w of length n is input to the model, after processing by RoBERTa-wwm and BiGRU, we get the final feature vector H = [H₁,H₂,⋯,H_n–1,H_n]. Then, for each layer of global pointers, the scoring function of global pointers is calculated as shown in Eq. (9), Eq. (10), and Eq. (11): (9) $q_{i, l} = W_{q, l} H_{i} + b_{q, l}$ \[{{q}_{i,l}}={{W}_{q,l}}{{H}_{i}}+{{b}_{q,l}}\] (10) $k_{i, l} = W_{k, l} H_{i} + b_{k, l}$ \[{{k}_{i,l}}={{W}_{k,l}}{{H}_{i}}+{{b}_{k,l}}\] (11) $s_{l} (i, j) = q_{i, l}^{T} k_{j, l}$ \[{{s}_{l}}(i,j)=q_{i,l}^{T}{{k}_{j,l}}\] where l represents the number of layers. q_i,l and k_i,l are the feature vectors H computed via the transformation matrices W_q,l and W_k,l and the offset indices b_q,l and b_k,l. s_l(i,j) represents the scoring function of fragment w_[ij] at layer l, i.e., the dot product of q_i,l and k_i,l, where w_[ij] represents the consecutive substring consisting of the ith through jth characters of text w, and fragment w_[ij], which satisfies s_l(i,j)>0, represents the output of the layer.

2.2.5

Loss function

This model will use the multicategorical cross entropy as the loss function, which is calculated as shown in Equation (12): (12) $L = \log (1 + \sum_{i \in P} e^{- S_{i}}) + \log (1 + \sum_{i \in N} e^{s_{i}})$ \[L=\log \left( 1+\sum\limits_{i\in P}{{{e}^{-{{S}_{i}}}}} \right)+\log \left( 1+\sum\limits_{i\in N}{{{e}^{{{s}_{i}}}}} \right)\] where P and N represent the set of positive and negative classes.

2.3

Relationship Extraction and Multimodal Entity Linking

2.3.1

Attention+BiLSTM based relation extraction

In this paper, five relationships are manually defined in the data schema stage: “inclusion relationship”, “procursive relationship”, “same relationship”, “brother relationship”, and “related relationship”. In this paper, algorithms, structures, and related terminology entities have been extracted from the text data. There is a wealth of semantic information between these entities, so this section needs to classify the relationship between the knowledge point entities and uncover the relationships between them.

2.3.2

Algorithm framework and process design

The main architecture of Attention+BiLSTM based classifier model contains five parts, which are input layer, embedding layer, LSTM network layer, attention layer, and output layer, and the Attention+BiLSTM model structure is shown in Figure 4. The embedding layer is to map each word into a vector, the BiLSTM layer is mainly used to obtain high-level features, the Attention layer calculates the weight of each temporal sequence, which is merged into a feature vector by weighting the vectors at each time step, and the output layer is to classify the learned feature vector into relationships. The structure of each layer is described next.

1)

Word Embedding

In this paper, a Word Embedding layer is added to the model to extract the lexical and syntactic features of entity pairs as inputs to the classifier model. The word vectors are represented by ids, and the syntactic features are formed by concatenating the ids of all the word vectors in a sentence to form a n*l-dimensional syntactic vector.

2)

Bidirectional Network

In this paper, BiLSTM is used to extend the unidirectional network by introducing one layer of LSTM, and the hidden connection to hidden connection of the two layers of LSTM is propagated in reverse order to connect to the next layer together. Thus, the model is able to utilize past and future information effectively. In this paper, the output vector of the hidden layer of the two-way LSTM is 64 dimensions. The vector output formula for the ind word is shown below: (13) $h_{i} = [\vec{h_{i}} \oplus {\bar{h}}_{i}]$ \[{{h}_{i}}=\left[ \overrightarrow{{{h}_{i}}}\oplus {{{\bar{h}}}_{i}} \right]\]

3)

Attention

The most important thing in this layer is the introduction of the weight parameter, which is actually a parameter that allocates attention, representing how much attention is allocated, and the size of the weight depends on the degree of contribution to the classification result. The weight parameter is learned from the sequence and finally the elements are merged according to their importance. The weight matrix is calculated by the following formula: (14) $M = \tanh (H)$ \[M=\tanh \left( H \right)\] (15) $α = s o f t \max (ω^{T} M)$ \[\alpha =soft\max \left( {{\omega }^{T}}M \right)\] (16) $r = H α^{T}$ \[r=H{{\alpha }^{T}}\] where H[h₁,h₂⋯h_T] is the set of all word vectors, a matrix consisting of the output vectors produced by the previous layer, where H∈R^{d^ω×T}. α is the result of the computation of the attention scores, ω is the parameter vector of the training, T represents the length of the sentence, and ω^T is the transpose of the parameter vector. r is the result of weighting H and the attention score. The final output state is obtained by tanh activation function, which is calculated as follows: (17) $h^{*} = \tanh (r)$ \[{{h}^{*}}=\tanh \left( r \right)\]

4)

Classifying

For the hidden state h^* obtained from the previous layer, the classifier takes it as input and uses softmax classifier to predict the label ŷ, which is calculated as follows: (18) $\hat{p} (y | S) = s o f t \max (W^{s} h^{*} + b^{(s)})$ \[\hat{p}\left( y\left| S \right. \right)=soft\max \left( {{W}^{s}}{{h}^{*}}+{{b}^{\left( s \right)}} \right)\] (19) $\hat{y} = \arg \max \sum \hat{p} (y | S)$ \[\widehat{y}=\arg \max \sum{{\hat{p}}}\left( y\left| S \right. \right)\]

2.4

Construction of a multimodal retrieval platform for courses based on knowledge graphs

Based on the constructed multimodal course knowledge graph, this paper designs and builds a multimodal retrieval platform for courses to better reflect the application value of this paper. The platform is based on the front-end framework Vue.js and the front-end visualization tool D3.js, and uses the application program interface Neo4j-Driver provided by Neo4j graph database to connect to the Neo4j database to realize the platform construction, and completes the graph search function based on the graph database Neo4j, as well as the visualization of multimodal knowledge map. The following section will introduce the overall architecture of the platform and demonstrate the application of the multimodal platform using a multimodal course knowledge graph retrieval case.Figure 5 depicts the overall architecture of the multimodal retrieval platform.

2.5

Precision teaching model based on knowledge graph learning platform

On the basis of related research and the knowledge graph learning platform’s support for accurate teaching, this paper builds an accurate teaching model based on knowledge graph learning platform based on theoretical foundations, and designs the various stages of the teaching model according to the principles of accurate teaching model based on knowledge graph learning platform, and builds an accurate teaching model based on knowledge graph learning platform. Precision teaching mode based on knowledge mapping learning platform Based on the typical information technology-supported precision teaching mode, the core and key links of precision teaching are clarified, the advantageous features are drawn upon, and the functional role of the knowledge mapping learning platform is relocated appropriately, so as to construct the precision teaching mode based on the knowledge mapping learning platform. This teaching model can realize the precision and personalization of teaching, which can significantly improve students’ English learning performance and increase the number of students with academic excellence and reduce the number of students with academic difficulties.

3

Results and discussion

In this section, the constructed teaching model based on knowledge mapping is applied in classroom practice to test its implementation effect, taking the English teaching classroom of a university as the object of empirical testing, the research subjects of this experiment are selected and analyzed through teaching practice, student questionnaires (40 questionnaires were issued and 40 were effectively returned) and teacher-student interviews, student performance, questionnaires and interviews are analyzed to deeply analyze the We analyzed the students’ performance, questionnaires and interviews to deeply analyze the effect of the teaching experiment and collect opinions and feedbacks, so as to further improve the design of English teaching mode in colleges and universities. In this paper, the object is divided into experimental and comparison classes, and a random distribution is used to eliminate interference from other factors.

3.1

Analysis of overall pre and post-test scores of experimental and control classes

The results of the paired samples t-test reflect that the mean of the post-test scores of both classes, the experimental and the control classes, are higher than the mean of the pre-test, and that there is a significant difference between the pre-test scores and the post-test scores. Therefore, the mean pre-test scores of the experimental class and the control class were compared and analyzed, and the significance of the improvement in academic performance during the experimental period was tested. At the same time, the rate of change in the number of people in each score band of the two classes was compared. Table 1 displays the statistics of the average pre- and post-test scores for the experimental and control classes.

Table 1.

The average performance statistics

	Laboratory class	Cross-reference class	variety
Premeasured mean	105.31	98.32	±6.99
Backmeasured mean	112.41	103.63	±8.78
variety	±7.1	±5.31

According to Table 1, it can be seen that the pre-test and post-test averages of the experimental class are higher than the pre-test averages of the control class, and the difference between the pre-test averages of the two classes is 6.99 points, and the difference between the post-test averages is 8.78 points, with 8.78>6.99, and the average change in the post-test scores of the experimental and control classes is higher than that of the pre-test scores. The mean values of the posttest of both the experimental and control classes were higher than the mean value of the pretest, the difference between the mean values of the pre and posttest of the experimental class was 7.1 points, and the difference between the mean values of the pre and posttest of the control class was 5.31 points, 7.1>5.31, and the magnitude of the change in the mean values of the pre and posttests of the experimental class was higher than the magnitude of the change in the mean values of the pre and posttests of the control class. Comparison of the difference in the change in mean scores indicates that the precise teaching model based on the knowledge graph learning platform has a positive effect on students’ academic performance, and the students in the experimental class have a higher rate of progress in academic performance than the control class.

In this study, the analysis of covariance was used to detect which teaching mode was more effective in improving students’ academic performance between the two. The analysis of covariance was conducted by considering the pre-test scores as a covariate, the experimental class and the control class as fixed factors, and the dependent variable as the post-test scores. The covariance analysis of the pretest academic performance of the experimental class and the control class is shown in Table 2. As can be seen from Table 2, the significant p-value of the pretest scores is 0.000, which is less than 0.05, indicating that the influence of the pretest scores on the posttest scores is more significant, so it is very necessary to set the pretest scores as a covariate. After excluding the confounding interference of the pretest scores, the significant P value of the teaching mode is 0.000, which is smaller than 0.05, indicating that different teaching modes present a significant impact on the posttest scores, and the mean value of the posttest scores of the experimental class is significantly higher than that of the control class, indicating that the teaching mode of the experimental class has a more significant impact on the students’ learning achievements. Therefore, the impact of the precise teaching mode based on the knowledge graph learning platform on students’ academic performance is higher than the impact of the original teaching mode on students’ academic performance.

Table 2.

The covariance analysis

source	Class III sum of squares	free degree	mean square	F	significance
Modified model	23001.855a	2	12032.889	127.010	0.000
intercept	1865.532	1	1784.632	22.316	0.000
pretest	22131.202	1	21156.233	251.322	0.000
Teaching model	1200.396	1	1211.022	12.237	0.000
error	7759.414	95	85.233
total	1242661.000	100
Revised total	32001.021	99

3.2

Analysis of overall satisfaction

The folding statistics of the satisfaction dimension are shown in Figure 6. The number of students who chose Agree and Strongly Agree is very high, indicating that the students are highly satisfied with the use of the English Knowledge Mapping Online Query Website system developed in this paper, and they think that the interface of the website is clear and concise, and the operation is intuitive and convenient. 16 out of 40 students think that “the interaction of the English Knowledge Mapping Online Query Website is good “, accounting for 40%. This indicates that the website needs to be improved in terms of interactivity and specific optimization needs to be done by conducting interviews with students.

As shown in Figure 7, 18 students, or 45%, agreed that “the website has the function of clearly displaying the knowledge structure”. Thirteen students agreed that “the website has efficient visual query function”, accounting for 33%. Eleven students agreed that the website has a comprehensive knowledge management function, accounting for 28%. This shows that the students agree that the English knowledge graphic interface and the English knowledge mapping query page meet their needs. It is also important to note that the students did not recognize the “comprehensive knowledge management functions” of the website, and most of them chose the option of “general” or “agree”. After consideration, the website concentrates on the fundamental functions of adding and saving English knowledge, which does not satisfy the students’ needs for comprehensive knowledge management.

The question in the questionnaire was to investigate the analysis of changes in students’ emotional experience and learning attitudes due to the website trigger. Learning attitude is an important psychological tendency mechanism that students play a role in the learning process, which can affect the learning effect. Based on the questionnaire data, the line graph of learning attitude dimensions is shown in Fig. 8. 23 out of 40 students agreed that “the website enhanced my interest in learning English”, accounting for 58% of the total number of students. Twenty-one students, or 53%, agreed that “the website has increased my confidence in learning English”. 20 students (50%) agreed that “the website has increased my motivation to learn English”. The website was cited by more than 60% of students as having increased their interest and confidence in English, stimulated their desire to learn, enhanced their initiative in learning, and gave them a good emotional experience.

The line graph of the learning effect dimension is shown in Figure 9. 21 out of 40 students (53%) agreed that “the website has enhanced my overall grasp of the structure of English knowledge”. 18 students (45%) agreed that “the website has enhanced my memorization of English knowledge”. Eighteen students, or 45%, agreed that “the website has enhanced my memorization of English knowledge”. Twenty students, or 50%, agreed that “the website has enhanced my ability to make knowledge connections”. This indicates that the learning method based on English knowledge mapping can increase students’ ability to grasp the overall structure of English and enhance their memory of knowledge, and is conducive to enhancing the ability to make connections between knowledge points, thus promoting the formation of their own English knowledge structure. However, there are some students who are not sure or do not agree that the knowledge mapping-based learning approach can enhance the ability to correlate their English knowledge points. This may be related to factors such as the short learning time and the first exposure of students to this learning approach.

4

Conclusion

The purpose of this paper is to propose a set of English teaching models that are based on knowledge mapping in the context of digital transformation and draw the following conclusions.

1) In the pre and post-test average scores of the experimental and control classes, the difference between the pre-test average of the experimental class and the control class is 6.99 points, the difference between the post-test average is 8.78 points, and 8.78>6.99, the difference between the pre and post-test averages of the experimental class is 7.1 points, the difference between the pre and post-test averages of the control class is 5.31 points, and 7.1>5.31, and a comparison of the difference in the change of average scores indicates that the knowledge graph-based Knowledge Graph Learning Platform’s precise teaching mode has a positive effect on students’ academic performance.

2) In the analysis of covariance between the pre-test and post-test scores of the experimental class and the control class, the significant P-value of the pre-test score is 0.000, which is smaller than 0.05, indicating that the pre-test score has a more significant effect on the post-test score, and the significant P-value of the teaching mode is 0.000, which is smaller than 0.05, indicating that different teaching modes present a significant effect on the post-test score, and the average value of the experimental class’s post-test score is significantly higher than the average value of the control class. The mean value is significantly higher than the mean value of the control class, and the influence of the English teaching mode based on knowledge mapping on students’ academic performance is more significant.

3) In the students’ satisfaction evaluation of English teaching mode, more than 60% of the students think that the English teaching mode based on knowledge mapping is conducive to enhancing confidence and interest in English learning, which indicates that the English teaching mode based on knowledge mapping has a good effect on English teaching.

The technical method proposed in this paper has been proven to effectively improve teaching quality, but the consideration of teaching efficiency is insufficient. Whether too complex technology can lead to a decline in teaching efficiency is an important direction for future research.

Language:: English

Publication timeframe:: 1 times per year
Journal Subjects:: Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other

Journal RSS Feed

Research on the Innovation of English Teaching Mode in Colleges and Universities under the Background of Digital Transformation

Bing Shen

Published Online: Mar 17, 2025

Received: Oct 23, 2024

Accepted: Feb 20, 2025

DOI: https://doi.org/10.2478/amns-2025-0246

KeywordsRoBERTa-wwm-BiGRU-GlobalPointer, English teaching model, Knowledge mapping, Relationship recognition

© 2025 Bing Shen, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Keywords
RoBERTa-wwm-BiGRU-GlobalPointer, English teaching model, Knowledge mapping, Relationship recognition