A Supervised Approach to Predict the Hierarchical Structure of Conversation Threads for Comments

Table 7

The difference between the evaluation metrics in presence and absence of features.

Feature	Dataset	Evaluation metric
		𝑃edge	𝑅edge	𝑃Path	𝑅Path	𝑃𝑅-path	𝑅𝑅-path	AccCTD
Similarity	ENENews	−0.00233	0.07415	−0.00538	0.03289	−0.01095	0.08996	0.04684
	Russianblog	0.00874	0.00012	0.00963	0.00347	0.00695	−0.01079	0.00355
Authors' language	ENENews	−0.01354	0.03246	−0.01785	0.00693	−0.02757	0.04052	0.00458
	Russianblog	−0.00229	0.01052	−0.00179	0.00750	−0.00590	0.00807	−0.00182
Global similarity	ENENews	−0.00402	0.05604	−0.01338	0.01581	−0.01129	0.07086	0.02742
	Russianblog	0.01090	0.01443	0.00582	0.00897	0.00368	−0.00317	−0.00087
Frequent words	ENENews	0.00252	0.00931	0.00170	0.00424	−0.00132	0.00973	0.00292
	Russianblog	0.00008	0.00023	0.00016	0.00029	0.00034	0.00047	−0.00011
Length ratio	ENENews	0.04206	−0.02984	0.04450	0.01700	0.06421	−0.05336	−0.00850
	Russianblog	0.00651	−0.00355	0.00920	0.00080	0.00784	−0.00592	0.00235
Authors' name	ENENews	0.03753	0.06392	0.01920	0.03285	0.02341	0.05998	0.01788
	Russianblog	0.00000	0.00000	0.00000	0.00000	0.00000	0.00000	0.00000
Frequent pattern	ENENews	−0.02028	0.02930	−0.02357	0.00300	−0.03793	0.03410	0.00526
	Russianblog	0.00458	0.06100	0.00830	0.05206	−0.00545	0.06150	0.00404
Location prior	ENENews	0.08778	−0.03690	0.08556	0.03280	0.11792	−0.07852	0.02925
	Russianblog	0.16091	0.09671	0.17766	0.13515	0.16699	0.08793	0.02414
Candidate filtering rule 1	ENENews	−0.05431	0.06138	−0.04102	0.01296	−0.06458	0.07545	0.06949
	Russianblog	−0.12506	0.34370	−0.13916	0.27445	−0.12683	0.37003	0.32546
Candidate filtering rule 2	ENENews	0.00014	−0.00004	0.00048	0.00041	−0.00038	−0.00035	−0.00009
	Russianblog	0.02575	0.02475	0.02462	0.02341	0.01746	0.00802	0.00000
Candidate filtering rule 3	ENENews	0.00357	−0.00121	0.00452	0.00254	0.00551	−0.00161	0.00161
	Russianblog	0.08521	−0.01978	0.09636	0.01702	0.09380	−0.03825	0.07371