Table 2.

Summary of the evaluation metrics and their respective interpretation.

Evaluation metric	Range of values	Definition	Interpretation
C_va	0 to 1	Measures how semantically similar words within a topic are to each other	A good and generally achievable range is 0.5<C_v<0.8.
Accuracy	0 to 1	Indicates the ratio between the number of correct predictions and the total number of predictions	Accuracy=1 indicates no incorrect predictions.
F1-score	0 to 1	The harmonic mean of a model’s ability to correctly predict positive instances (recall) and minimize predicting negative instances as positive (precision)	F1-score=1 indicates that the model perfectly predicts positive instances and does not misclassify negative instances as positive.
MCCb	−1 to 1	Measures the relationship between the number of positive instances correctly classified, the number of negative instances correctly classified, and the number of positive and negative instances misclassified	MCC=−1 indicates a perfectly inaccurate classifier. MCC=0 indicates a perfectly random classifier. MCC=1 indicates a perfectly accurate classifier.
Cohen κ	0 to 1	Measures the interrater agreement of 2 raters (in machine learning, this is the classifier and the ground truth)	Cohen κ=0 indicates no agreement. Cohen κ=1 indicates perfect agreement.
Brier Loss	0 to 1	A cost function that measures the difference between the predicted probability and the ground truth	Brier Loss=0 indicates perfect accuracy. Brier Loss=1 indicates perfect inaccuracy.

^aC_v: coherence score.

^bMCC: Matthews correlation coefficient.