Table 3.

Long COVID diagnosis: recently applied data mining and NLP techniques.

Study	Input data	AI method	Task	Output (%)
Miao et al. (2022)	Tweets	NLP	Analysis of reported Long COVID symptoms in terms of demographics, geographical and temporal parameters	Accuracy demographic categories 89 symptom categories 95
Zhu et al. (2022)	Clinical notes	Pretrained BERT	Identification of Long COVID and potential computational phenotypes	Sensitivity score 88.1
Scarpino et al. (2022)	Blogs	LDA and BERT	Extract discussion topics in the Italian narration of COVID-19 pandemic	Accuracy of BERT 91.97
Matharaarachchi et al. (2022)	Tweets	Association rule mining	Relationships between symptoms	Confidence 77 for lung/breathing problems and loss of taste vs. loss of smell
Wang et al. (2022)	Clinical notes	PASCLex (NLP) model	Identification of symptoms	Precision 94 recall 84
Banda et al. (2021)	Tweets	NLP and SVM	Identification of symptoms	Accuracy 75 on a 20% random held-out test set
Déguilhem et al. (2022)	Tweets	Biterm Topic Modeling	Identification and co-occurrence of symptoms	Three major symptom co-occurrences: asthenia-dyspnea 35.3, asthenia-anxiety 22.5, asthenia-headaches 17.3

The division is between BERT (the first 3 rows) and other approaches (all reported measures have the same number of decimal digits as the original paper).