Open Access

Curriculum Content Construction and Updating for Intelligent Educational Systems Using Knowledge Graphs

  
Sep 26, 2025

Cite
Download Cover

Figure 1.

The guide flow chart of the building d3.js force
The guide flow chart of the building d3.js force

Figure 2.

The changes of loss rate
The changes of loss rate

Figure 3.

The F1 value of training
The F1 value of training

Figure 4.

The comparison of the deep learning model
The comparison of the deep learning model

Figure 5.

The changes of loss rate
The changes of loss rate

Figure 6.

The F1 value of training
The F1 value of training

Figure 7.

The comparison of the effects of multiple models
The comparison of the effects of multiple models

Comparison results

No. Number of questions The number of answers returned The correct number of answers Accuracy Precision
1 120 108 95 79.2% 87.96%
2 120 103 89 74.2% 86.41%
3 120 91 84 70.0% 92.31%
4 120 96 85 70.8% 88.54%
5 120 106 91 75.8% 85.85%
6 120 110 97 80.8% 88.18%
7 120 107 96 80.0% 89.72%
8 120 97 90 75.0% 92.78%
9 120 105 98 81.7% 93.33%
10 120 89 82 68.3% 92.13%
Average 120 101.2 90.7 75.6% 89.62%

Test results of different models

Model Precision (%) Recall (%) F1 (%)
CRF 73.22 66.08 69.47
Bi_LSTM 82.28 75.23 78.60
Bi_LSTM+CRF 87.35 87.63 87.49
BERT 94.69 95.33 95.01
Ours 98.87 98.98 98.93

Experimental comparison of various models (%)

Sorting algorithm Macro accuracy Macro recall Macro F1
SVM 87.36 85.81 86.58
KNN 86.84 89.38 88.09
Ours 91.54 92.22 91.88

Physical identification of experimental parameters

Experimental parameters Value
train_batch_size 48
eval_batch_size 64
max_seq_length 64
num_train_epochs 10
drop_out 0.5
embedding_size 752
learning_rate 5.0×10-5
hidden_dropout_prob 48

Classifier experimental effect (%)

Question type Precision Recall F1
Factual type 92.22 92.63 92.42
Statistical type 92.63 93.48 93.05
Whether or not type 91.82 92.54 92.18
List type 91.51 91.29 91.40
Method type 89.61 90.17 89.89
Language:
English