Table 5. Five-class sentiment classification studies.

Study	Dataset	Model	Classification metrics
			Accuracy	Precision	Recall	${\mathbf{F}}_{\mathbf{1}}^{\mathbf{m}\mathbf{a}\mathbf{c}\mathbf{r}\mathbf{o}}$	${\mathbf{F}}_{\mathbf{1}}^{\mathbf{w}\mathbf{e}\mathbf{i}\mathbf{g}\mathbf{h}\mathbf{t}\mathbf{e}\mathbf{d}}$
Muhammad et al. (2022)	NaijaSenti	XLM-R-base+LAFT	n/a	n/a	n/a	n/a	0.795
Muhammad et al. (2022)	NaijaSenti	M-BERT+LAFT	n/a	n/a	n/a	n/a	0.7700
Fiok et al. (2021)	sentiment@USNavy	BART large + CNN	n/a	n/a	n/a	0.596	n/a
Smetanin & Komarov (2021a)	RuSentiment	M-BERT-Base	n/a	0.6722	0.6907	0.6794	0.7244
Smetanin & Komarov (2021a)	RuSentiment	RuBERT	n/a	0.7089	0.7362	0.7203	0.7571
Smetanin & Komarov (2021a)	RuSentiment	M-USE-CNN	n/a	0.6571	0.6708	0.6627	0.7105
Smetanin & Komarov (2021a)	RuSentiment	M-USE-Trans	n/a	0.6821	0.6982	0.6860	0.7342
Jamadi Khiabani, Basiri & Rastegari (2020)	TripAdvisor	Dempster–Shafer-based model	0.79	0.5	0.47	0.49	n/a
Jamadi Khiabani, Basiri & Rastegari (2020)	CitySearch	Dempster–Shafer-based model	0.79	0.48	0.48	0.48	n/a
Kuratov & Arkhipov (2019)	RuSentiment	Multilingual BERT	n/a	n/a	n/a	n/a	0.7082
Kuratov & Arkhipov (2019)	RuSentiment	RuBERT	n/a	n/a	n/a	n/a	0.7263
Baymurzina, Kuznetsov & Burtsev (2019)	RuSentiment	SWCNN + fastText Twitter	n/a	n/a	n/a	n/a	0.7850
Baymurzina, Kuznetsov & Burtsev (2019)	RuSentiment	BiGRU + ELMo Wiki	n/a	n/a	n/a	n/a	0.6947
Tripto & Ali (2018)	YouTube	LSTM	0.5424	n/a	n/a	0.5320	n/a
Li et al. (2018)	Twitter	Logistic Regression	0.6899	0.6053	0.6899	0.6354	n/a
Ahmadi et al. (2017)	SST-5	RNTN	0.41	n/a	n/a	0.32	n/a
Buntoro, Adji & Purnamasari (2016)	Twitter	Naïve Bayes	0.7177	0.716	0.718	n/a	n/a
Aly & Atiya (2013)	LABR	SVM	0.503	n/a	n/a	n/a	0.491
Chetvirokin & Loukachevitch (2013)	ROMIP-2012 (Movies)	n/a	0.407	n/a	n/a	0.377	n/a
Blinov, Kotelnikov & Pestov (2013)	ROMIP-2012 (Books)	SVM	0.481	0.339	0.496	0.402	n/a
Chetvirokin & Loukachevitch (2013)	ROMIP-2012 (Cameras)	n/a	0.480	n/a	n/a	0.336	n/a
Pak & Paroubek (2012)	ROMIP-2011 (Movies)	SVM	0.599	n/a	n/a	0.286	n/a
Pak & Paroubek (2012)	ROMIP-2011 (Books)	SVM	0.622	n/a	n/a	0.291	n/a
Pak & Paroubek (2012)	ROMIP-2011 (Cameras)	SVM	0.626	n/a	n/a	0.342	n/a

Note:

We selected only those studies, which consideredfive sentiment classes and reported at least one of the following classification measures: Precision, Recall, macro F1, weighted F1. Among all datasets, only ROMIP (Chetviorkin, Braslavskiy & Loukachevich, 2013; Chetvirokin & Loukachevitch, 2013) and RuSentiment (Rogers et al., 2018) datasets are in Russian.