Abstractive Arabic Text Summarization Based on Deep Learning

Table 4

The result of applying GRU, LSTM, and BiLSTM with different layers at the encoder to the AMN dataset.

Model	No. of layers	ROUGE-1			ROUGE-2			ROUGE-L			BLEU
		Precision	Recall	F1	Precision	Recall	F1	Precision	Recall	F1
GRU	1	19.89	17.81	18.17	1.28	1.14	1.16	13.62	12.88	12.11	0.22
	2	39.04	30.77	33.13	8.66	7.15	7.62	25.05	21.61	20.96	0.35
	3	35.31	26.81	29.23	6.33	5.2	5.54	22.14	18.57	18.1	0.32
LSTM	1	39.41	32.66	34.78	11.47	9.91	10.38	26.76	23.52	23.28	0.34
	2	43.47	36.23	38.2	14.5	12.71	13.28	31.13	27.01	27.14	0.36
	3	36.26	32.35	33.34	10.36	9.61	9.76	25.68	23.66	23.05	0.349
BiLSTM	1	48.41	41.38	43.67	18.99	17.14	17.72	34.8	31.59	31.37	0.39
	2	43.6	32.12	35.75	10.39	8.37	9.01	27.46	22.71	22.73	0.34
	3	48.15	42.65	44.28	19.46	17.93	18.35	35.48	32.86	32.46	0.41

The highest result is given in bold.