An Intelligent Classification Method for Online Resource Data of College Language Teaching Based on Deep Reinforcement Learning

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Evaluation for the college Chinese online teaching resource classification of DRML model

Index	Evaluation score (percentage)
	-2	-1	0	1	2	1/2	Mean
Classification accuracy	1.79%	4.91%	12.14%	54.76%	26.40%	81.16%	0.99
Resource quality	0.19%	2.31%	10.45%	55.82%	31.23%	87.05%	1.16
Classification speed	0.38%	4.14%	12.42%	53.69%	29.37%	83.06%	1.08
Result clarity	0.97%	2.45%	12.36%	51.28%	32.94%	84.22%	1.13
Learning difficulty	1.72%	2.47%	20.40%	57.05%	18.36%	75.41%	0.88
Acquisition difficulty	0.18%	2.74%	21.58%	53.32%	22.18%	75.50%	0.95
Efficiency improvement	1.32%	2.07%	15.79%	57.07%	23.75%	80.82%	1.00
Tool efficiency	1.64%	4.80%	16.60%	46.04%	30.92%	76.96%	1.00
Adopt willingness	1.18%	3.27%	11.36%	47.99%	36.20%	84.19%	1.15
Using willingness	0.39%	1.87%	11.89%	54.68%	31.17%	85.85%	1.14

Ablation experiment results (%)

Dataset	P@k	Static	Dynamic	Difference
Eurlex-4k	P@1	85.46	89.63	+4.17
	P@3	74.59	78.45	+3.86
	P@5	66.32	67.96	+1.64
AmazonCat-13k	P@1	84.26	87.58	+3.32
	P@3	73.94	75.68	+1.74
	P@5	67.06	69.33	+2.27
Wiki10-31k	P@1	87.16	89.56	+2.40
	P@3	76.49	80.02	+3.53
	P@5	69.28	70.63	+1.35

Relationship between emotion, label semantic features and classification performance (%)

Dataset	P@k	Emotion feature	Label semantic feature	Difference
Eurlex-4k	P@1	85.95	89.42	+3.47
	P@3	76.42	79.12	+2.70
	P@5	66.37	68.15	+1.78
AmazonCat-13k	P@1	94.86	98.43	+3.57
	P@3	84.25	86.44	+2.19
	P@5	68.36	69.75	+1.39
Wiki10-31k	P@1	89.45	91.64	+2.19
	P@3	79.64	81.06	+1.42
	P@5	69.63	70.87	+1.24

Comparison of classification accuracy of different methods (%)

Method	miniImageNet	tieredImageNet	QHGIM
ProroNet	69.48	75.46	73.52
RelationNet	70.56	75.08	70.63
SimCLR	81.03	80.12	76.28
SimSiam	81.89	83.44	78.49
TPMN	85.65	85.47	80.36
RE-Net	84.74	84.23	80.07
ProroNet+Swin	75.64	78.55	74.68
BML	77.42	84.64	83.59
SUN	85.79	87.09	83.66
DRML	88.96	90.41	89.73

Comparison experiment of DRML and other reference models (%)

Algorithm	Eurlex-4k			AmazonCat-13k			Wiki10-31k
	P@1	P@3	P@5	P@1	P@3	P@5	P@1	P@3	P@5
PfastreXML	74.91	73.45	57.02	92.31	78.04	63.56	82.84	69.88	60.11
DisMec	83.90	72.02	61.15	97.46	81.12	67.48	85.79	74.92	64.88
Parabel	81.41	70.49	55.96	94.05	78.89	64.11	85.89	75.67	59.42
SLEEC	88.01	71.57	58.40	96.69	79.84	61.33	86.40	79.60	69.35
XML-CNN	83.18	65.77	54.59	91.73	82.04	66.88	87.02	75.89	61.37
LAHA	80.37	69.04	58.53	94.92	80.16	65.78	85.24	76.39	62.87
AttentionXML	81.86	77.95	62.62	95.13	85.16	66.60	85.94	79.63	60.45
X-Transformer	74.34	78.36	62.51	93.60	83.38	69.18	88.12	80.70	70.51
APLC-XLNet	81.14	68.85	58.15	97.25	79.49	69.54	84.75	77.78	64.17
LightXML	89.39	66.34	64.22	94.58	86.73	65.18	89.86	81.14	64.82
DRML	90.25	80.56	69.74	98.71	86.47	71.22	91.04	82.43	71.53