Skip to main content

Table 2 Cluster evaluation on AG News dataset

From: The performance of BERT as data representation of text clustering

Method

AG news

ACC

NMI

ARI

TFIDF  +  KM

0.5019  ±  0.0718

0.2559 ± 0.0802

0.2552 ± 0.0803

BERT + Max + I + KM

0.7674 ± 0.0018

0.4872 ± 0.0021

0.4868 ± 0.0021

BERT + Max + LN + KM

0.7913 ± 0.0040

0.5199 ± 0.0050

0.5195 ± 0.0050

BERT + Max + N + KM

0.7858 ± 0.0017

0.5136 ± 0.0025

0.5132 ± 0.0025

BERT + Max + MM + KM

0.4408 ± 0.0012

0.1986 ± 0.0014

0.1979 ± 0.0014

BERT + Mean + I + KM

0.6491 ± 0.0016

0.4196 ± 0.0010

0.4191 ± 0.0010

BERT + Mean + LN + KM

0.6468 ± 0.0036

0.4152 ± 0.0018

0.4148 ± 0.0018

BERT + Mean + N + KM

0.6467 ± 0.0033

0.4151 ± 0.0017

0.4146 ± 0.0017

BERT + Mean + MM + KM

0.3208 ± 0.0051

0.0441 ± 0.0008

0.0432 ± 0.0008

TFIDF + EFCM

0.5788 ± 0.03197

0.2979 ± 0.0309

0.2973 ± 0.0309

BERT + Max + I +  EFCM

0.7561 ± 0.0004

0.4731 ± 0.0006

0.4726 ± 0.0006

BERT + Max + LN +  EFCM

0.778 ± 0.0002

0.4976 ± 0.0004

0.4972 ± 0.0004

BERT + Max + N +  EFCM

0.7642 ± 0.0003

0.4841 ± 0.0004

0.4837 ± 0.0004

BERT + Max + MM +  EFCM

0.4439 ± 0.0085

0.1997 ± 0.0100

0.1991 ± 0.0100

BERT + Mean + I +  EFCM

0.6449 ± 0.0003

0.4086 ± 0.0002

0.4081 ± 0.0002

BERT + Mean + LN +  EFCM

0.6423 ± 0.0003

0.4088 ± 0.0003

0.4083 ± 0.0003

BERT + Mean + N +  EFCM

0.6425 ± 0.0003

0.4089 ± 0.0003

0.4084 ± 0.0003

BERT + Mean + MM +  EFCM

0.3067 ± 0.0037

0.0429 ± 0.0003

0.0421 ± 0.0003

TFIDF + DEC

0.7211 ± 0.0250

0.3861 ± 0.0265

0.4139 ± 0.0338

BERT + Max + I +  DEC

0.2539 ± 0.0274

0.0037 ± 0.0259

0.003 ± 0.0210

BERT + Max + LN +  DEC

0.7677 ± 0.0436

0.4878 ± 0.0344

0.513 ± 0.0483

BERT + Max + N +  DEC

0.2585 ± 0.0326

0.004 ± 0.0179

0.0033 ± 0.0162

BERT + Max + MM +  DEC

0.3529 ± 0.1505

0.0817 ± 0.1476

0.0798 ± 0.1461

BERT + Mean + I +  DEC

0.7719 ± 0.0506

0.5055 ± 0.0363

0.5304 ± 0.0518

BERT + Mean + LN +  DEC

0.7653 ± 0.0550

0.4987 ± 0.0426

0.5206 ± 0.0579

BERT + Mean + N +  DEC

0.8038 ± 0.0325

0.538 ± 0.0210

0.5707 ± 0.0296

BERT + Mean + MM +  DEC

0.25 ± 0

0.0004 ± 0.0018

0 ± 0

TFIDF + IDEC

0.7453 ± 0.0243

0.4251 ± 0.0244

0.4571 ± 0.0315

BERT + Max + I +  IDEC

0.376 ± 0.1413

0.1467 ± 0.1565

0.1253 ± 0.1457

BERT + Max + LN +  IDEC

0.7819 ± 0.0411

0.5131 ± 0.0294

0.5394 ± 0.0428

BERT + Max + N +  IDEC

0.3618 ± 0.1478

0.1163 ± 0.1511

0.1072 ± 0.1408

BERT + Max + MM +  IDEC

0.4077 ± 0.111

0.1157 ± 0.1269

0.1093 ± 0.1222

BERT + Mean + I +  IDEC

0.7836 ± 0.0509

0.5296 ± 0.0353

0.5544 ± 0.0511

BERT + Mean + LN +  IDEC

0.782 ± 0.0541

0.5297 ± 0.0398

0.5524 ± 0.0548

BERT + Mean + N +  IDEC

0.8019 ± 0.0330

0.5383 ± 0.0217

0.5688 ± 0.0312

BERT + Mean + MM +  IDEC

0.2616 ± 0.0208

0.0165 ± 0.0184

0.0026 ± 0.0063

  1. The feature extraction and normalization strategies are abbreviated into Max for max pooling, Mean for mean pooling, I for identity normalization, LN for layer normalization, N for standard normalization, and MM for min–max normalization. The deviations denote the standard deviation of the metric from 50 repetitions. The values in bold denote the highest value in every metric in each text clustering algorithm. While the methods in bold, if there are any, is the best performing method in each text clustering algorithm