Intermedia Agenda Setting amid the Pandemic: A Computational Analysis of China’s Online News

Table 1

Extended metrics for evaluating the performance of classifiers constructed by machine learning for the COVID-19 pandemic news coverage.

Topics	F1-score	Precision	Recall
Pandemic notifications	0.93	0.94	0.93
Treatment on patients	0.75	0.78	0.73
Methods for pandemic prevention	0.78	0.75	0.81
Scientific knowledge	0.81	0.79	0.83
Social assistance	0.69	0.80	0.60
Work resumption	0.85	0.82	0.87
International pandemic situation	0.90	0.89	0.92
Propaganda and mobilization	0.80	0.77	0.83

Note. We divided all the manually labeled Weibo posts into a training set and a testing set. Two tasks were taken for an SVM model. For the first task, we used the Weibo posts in the training set to build an SVM model. For the second task, we used the SVM model to predict the topic category in the testing set. To evaluate the performance of supervised machine learning, we compared the SVM-predicted labels and the manual labels by generating the precision and recall values, where and [51]. The prediction value measures how many of all the tweets predicted by the SVM model as this specific topic were indeed the same topic coded by manual coding. The recall value measures how many of all the tweets coded as this specific topic by manual coding were indeed predicted by the SVM model. Finally, an F-score is the weighted average of prediction and recall.