Deep Learning-Based Methods for Sentiment Analysis on Nepali COVID-19-Related Tweets

Table 1

Summary performance (%) of sentiment classification of some state-of-the-art methods on tweets.

Method	Dataset	Accuracy (%)	Methodology	Limitations
Basir et al. [8]	StanfordSentiment140 [8]	85.4	(i) Stacked ensemble method	(i) Ignore the effects of global COVID-19 news on sentiment analysis beside the specific country
Rustam et al. [7]	Covid-19Tweets [20]	93.0	(ii) BoW with various ML methods	(ii) Limited performance on small datasets
Aljameel et al. [14]	Self-created dataset	85.0	(iii) N-gram with various ML methods	(iii) Feature selection and hyperparameter tuning operation is not performed
Ramya et al. [21]	Self-created dataset	91.0	(iv) Naive Bayes	(iv) Experimented with a limited data
Naseem et al. [5]	COVIDSenti [5]	92.2	(v) ML methods such as Naive Bayes, support vector machine, and random forest	(v) Limited to English tweets
Satu et al. [10]	Covid-19Tweets [20]	98.8	(vi) Cluster-based classification	(vi) Tweets search limited to a few keywords in English text