Improving topic modeling through homophily for legal documents

Applied Network Science

Table 3 Coherence, Cv, UMass, and \(C_{UCI}\) scores of topics on \(G'\) against cutting and weighting models

Model	\(G', w_n, w_j\)	Coherence	Cv	UMass	\(C_{UCI}\)
LDA		0.131	0.509	\(-\) 0.457	\(-\) 4.476
RTM	\(w \rightarrow \infty\)	0.159	0.611	\(-\) 0.412	\(-\) 1.078
Cutting	\(w_n^P \, \ge\) 0, \(w_n^S \, \ge\) 0	0.163	0.616	\(-\) 0.394	\(-\) 1.061
	\(w_n^P \, \ge\) 5, \(w_n^S \, \ge\) 10	0.163	0.619	\(-\) 0.395	\(-\) 1.029
	\(w_j^P \, \ge\) 0.50, \(w_j^S \, \ge\) 0.25	0.167	0.613	\(-\) 0.402	\(-\) 1.057
Weighting	\(w_j^P + w_j^S\)	0.165	0.620	\(-\) 0.387	\(-\) 1.031
	\(w_j^P \times w_j^S\)	0.174	0.610	\(-\) 0.419	\(-\) 1.052
	\(T_{H,0}(w_j^P, w_j^S)\)	0.173	0.607	\(-\) 0.416	\(-\) 1.083
	\(T_{Y,1}(w_j^P, w_j^S)\)	0.174	0.622	\(-\)0.371	\(-\) 1.030
	\(T_{E}(w_j^P, w_j^S)\)	0.171	0.614	\(-\) 0.395	\(-\) 1.030
	\(w_s^P + w_s^S\)	0.172	0.619	\(-\) 0.387	\(-\) 1.020
	\(w_s^P \times w_s^S\)	0.170	0.614	\(-\) 0.392	\(-\) 1.050
	\(T_{H,0}(w_s^P, w_s^S)\)	0.171	0.613	\(-\) 0.402	\(-\) 1.082
	\(T_{Y,1}(w_s^P, w_s^S)\)	0.171	0.617	\(-\) 0.399	\(-\) 1.055
	\(T_{E}(w_s^P, w_s^S)\)	0.171	0.620	\(-\) 0.387	\(-\)1.012