Table 5 Ablation experiment results (percentage) of the DNCL model. ’w/o DG’ represents the DNCL model without the dual-gate mechanism, ‘w/o AN’ represents the DNCL model without the adversarial noise modeling mechanism, ’w/o intra’ represents the DNCL model without use the intra-layer contrastive learning, ’w/o inter’ represents the DNCL model without use inter-layer contrastive learning and ‘w/o CL’ represents the DNCL model without use the multi-layer embedding contrastive learning strategy.

Model	ICEWS14				ICEWS05-15				ICEWS18				GDELT
Model	MRR	H@1	H@3	H@10	MRR	H@1	H@3	H@10	MRR	H@1	H@3	H@10	MRR	H@1	H@3	H@10
DNCL	51.18	40.37	56.86	71.89	59.02	48.06	65.50	80.88	37.43	25.83	42.70	60.01	23.59	14.78	25.40	42.39
w/o DG	50.05	38.87	56.18	71.70	58.70	47.62	65.34	80.04	36.45	24.88	41.74	58.89	21.69	12.49	23.53	40.34
w/o AN	50.27	38.90	56.30	71.67	58.82	47.78	65.46	79.91	37.03	25.69	42.10	59.16	21.11	11.97	22.79	39.62
w/o intra	50.19	38.81	56.18	72.24	58.70	47.92	65.13	79.51	37.17	25.78	42.59	59.86	23.09	13.98	25.06	42.17
w/o inter	50.40	39.10	56.71	72.11	58.65	47.89	65.19	79.84	36.83	25.33	41.89	59.42	23.35	14.12	25.18	42.33
w/o CL	47.50	35.79	53.82	70.72	58.41	47.37	64.99	79.72	36.77	25.10	41.93	59.63	22.89	13.78	24.55	41.53

The best result is highlighted in black font and the second best result is underlined.
The model results with the best performance under each evaluation indicator.

Quick links

Search