Aspect-Level Sentiment Analysis Approach via BERT and Aspect Feature Location Model

Table 4

Some examples of aspect-based sentiment analysis.

Word embedding	Models	Laptop		Restaurant		Twitter
		Accuracy	Macro-F1	Accuracy	Macro-F1	Accuracy	Macro-F1
Embedding	LSTM	0.6144	0.4401	0.7304	0.533	0.6474	0.6058
	ATAE-LSTM	0.6019	0.4909	0.7375	0.5725	0.6864	0.6501
	Cabasc	0.6301	0.5297	0.7241	0.5245	0.6171	0.5657
	IAN	0.6191	0.4671	0.7268	0.4897	0.6618	0.6251
	MGAN	0.5878	0.4264	0.7179	0.4974	0.6373	0.5856
	MemNet	0.7915	0.7576	0.8241	0.7313	0.6936	0.6772
	RAM	0.5956	0.4308	0.7152	0.4656	0.6358	0.5998
	TNet	0.7022	0.6404	0.7688	0.6269	0.6994	0.6827
BERT	IAN-BERT	0.7696	0.7179	0.808	0.722	0.7269	0.7048
	AOA-BERT	0.7774	0.7407	0.7341	0.7173	0.7341	0.7173
	MemNet-BERT	0.7915	0.7576	0.8241	0.7313	0.6936	0.6772
	AEN-BERT	0.7947	0.7544	0.8125	0.7069	0.7168	0.7052
	BERT-base	0.768	0.7288	0.8375	0.7613	0.7442	0.7271
	LCF-bert	0.7837	0.7441	0.8509	0.7894	0.7254	0.7113
	ALM-BERT	0.8009	0.7603	0.8545	0.795	0.7413	0.73