Modelling the Latent Semantics of Diffusion Sources in Information Cascade Prediction

Table 2

Evaluation results on Twitter and Weibo datasets.

Method/gain	Twitter				Weibo
	MRR	Hit@10	Hit@50	Hit@100	MRR	Hit@10	Hit@50	Hit@100
RNN	21.37	23.82	25.93	27.19	1.76	2.82	8.09	13.17
RNN + DSSF	23.03	26.77	31.81	33.55	1.89	3.00	8.73	14.05
Gain	8.04	12.38	22.68	23.39	7.38	6.38	7.91	6.68
LSTM	21.82	26.17	28.45	29.41	1.38	2.74	8.89	14.59
LSTM + DSSF	23.51	28.59	33.25	35.11	1.67	3.15	9.73	15.75
Gain	7.75	9.25	16.87	19.38	21.01	14.96	9.45	7.95
DeepDiffuse	18.34	25.93	28.63	29.89	1.61	3.06	9.78	15.84
DeepDiffuse + DSSF	18.92	26.97	32.41	34.45	1.89	3.37	10.21	17.27
Gain	3.16	4.01	13.20	15.25	17.39	10.13	4.39	9.02
HiDAN	22.34	25.87	28.51	29.41	1.90	3.54	10.03	15.86
HiDAN + DSSF	23.11	27.67	32.11	34.84	2.11	3.83	11.31	17.12
Gain	3.46	6.95	12.62	18.46	11.05	8.19	12.76	7.94