SentMask: A Sentence-Aware Mask Attention-Guided Two-Stage Text Summarization Component

Table 2

Details of the ROUGE and BLEU evaluation values in the baseline models on the MS2 dataset.

Models	RG1	RG2	RGL	BE	BE1	BE2	BE3	BE4
Semisupervised
Lead3	25	10.5	21.78	4.77	12.12	5.76	3.47	2.14
TextRank	18.54	5.58	16.03	2.88	10.29	3.38	1.83	1.07
GenCompareSum	29.83	14.22	24.96	7.71	14.15	7.22	2.99	2.99
TextRank + Seq2seq	24.58	10.78	20.15	6.22	12.33	9.28	3.29	2.09
SentMask	46.36	24.56	41.81	16.35	39.47	22	13.19	8.04
Supervised
Seq2Seq	34.45	19.25	31.64	8.27	15.67	9.54	6.02	4.41
Pointer-gen	35.34	16.28	31.43	7.63	18.31	9.08	5.64	3.62
Global encoding	29.67	14.53	24.34	12.54	30.1	15.01	10.32	6.17
Presumm	35.99	16.88	30.72	13.69	32.4	16.32	10.35	6.41
BART	52.97	33.41	49.15	28.85	54.22	34.83	25.64	19.41
SentMask	55.38	35.97	51.8	30.94	55.51	36.75	27.69	21.48

The best values in the metric are in bold. RG1: ROUGE-1; RG2: ROUGE-2; RGL: ROUGE-L; BE: BLEU; BE1: BLEU1; BE2: BLEU2; BE3: BLEU3; BE4: BLEU4.