MIT Press

Skip to Main Content

Table 2:

C_v coherence scores for latent topics produced by different models. The best result in each column is highlighted in bold. Our joint model Topic+Disc achieves significantly better coherence scores than all the baselines (p < 0.01, paired test).

Models	K = 50		K = 100
	TREC	TWT16	TREC	TWT16
Baselines
LDA	0.467	0.454	0.467	0.454
BTM	0.460	0.461	0.466	0.463
LF-DMM	0.456	0.448	0.463	0.466
LF-LDA	0.470	0.456	0.467	0.453
NTM	0.478	0.479	0.482	0.443
Li et al. (2018)	0.463	0.433	0.464	0.435
Our models
Topic only	0.478	0.482	0.481	0.471
Topic+Disc	0.485	0.487	0.496	0.480

Models	K = 50		K = 100
	TREC	TWT16	TREC	TWT16
Baselines
LDA	0.467	0.454	0.467	0.454
BTM	0.460	0.461	0.466	0.463
LF-DMM	0.456	0.448	0.463	0.466
LF-LDA	0.470	0.456	0.467	0.453
NTM	0.478	0.479	0.482	0.443
Li et al. (2018)	0.463	0.433	0.464	0.435
Our models
Topic only	0.478	0.482	0.481	0.471
Topic+Disc	0.485	0.487	0.496	0.480