Table 4.

Participating systems in subtasks. A total of 19 participating systems and 2 baseline systems are constructed for the Japanese subtask, 12 participating systems and 2 baseline systems are constructed for the English subtask, and 6 participating systems and 2 baseline systems are constructed for the Chinese subtask.

System ID	Models or methods	Language resources
AITOK-ja [33]	Keyword-based, logistic regression, and SVMa,b	—c
AKBL-ja and AKBL-en [34]	SVM and Fisher exact test	Patient symptom feature word dictionary and Disease-X feature words dict1 and dict2
DrG-ja [35]	Random forest	—
KIS-ja [36]	Rule-based and SVM	—
NAIST-ja, NAIST-en, and NAIST-zh [37]	Ensembles of hierarchical attention network and deep character-level convolutional neural network with loss functions (negative loss function, hinge, and hinge squared)	—
NIL-ja [38]	Rule-based	—
NTTMU-ja [39]	Principle-based approach	Manually constructed knowledge for capturing tweets that conveyed flu-related information, using common sense and ICD-10d
NTTMU-en [39]	SVM and recurrent neural network	Manually constructed knowledge for capturing tweets that conveyed flu-related information, using common sense and ICD-10
TUA1-zh [40]	Logistic regression, SVM, and logistic regression with semantic information	Updated training samples using active learning unlabeled posts downloaded with the symptom names in Chinese
UE-ja [41]	Rule-based and random forest	Custom dictionary consisting of nouns selected from the dry-run dataset and heuristics
UE-en [41]	Rule-based, random forests, and skip-gram neural network for word2vec	Custom dictionary consisting of nouns selected from the dry-run dataset and heuristics
Baseline	SVM (unigram and bigram)	—

^aSVM: support vector machine.

^bIt indicates that the method was tested after the submission of the formal run, and thus, it was not included in the results.

^cIt indicates that any language resources were not used.

^dICD: International Codes for Diseases.