Table 3. The quantitative results for accuracy, precision, and recall of SparkText using three datasets.

For each dataset, 80% was used to train a prediction model and the remaining 20% for testing.

Dataset	Classifier	Accuracy	Precision	Recall
Abstracts	SVM	94.63%	93.11%	94.81%
Abstracts	Logistic Regression	92.19%	91.07%	89.49%
Abstracts	Naïve Byes	89.38%	89.13%	90.82%
Full-text Articles I	SVM	94.47%	92.97%	93.14%
Full-text Articles I	Logistic Regression	91.05%	90.77%	89.19%
Full-text Articles I	Naïve Bayes	88.02%	89.01%	90.68%
Full-text Articles II	SVM	93.81%	91.88%	92.27%
Full-text Articles II	Logistic Regression	90.57%	90.28%	91.59%
Full-text Articles II	Naïve Bayes	86.44%	87.61%	89.12%