3. Results and discussion

3.1 Performance evaluation

In order to show the model performance on the reaction superiority classification task, the accuracy, precision, recall, F1-score and the area under receiver operating characteristic curve (ROC-AUC) are adopted as evaluation metrics. In this work, the developed model is compared with other models by examining their performance with different message aggregation methods and investigating the utilization of the AIR residual in exploring the appropriate structure in the backbone model. The testing set evaluation results of all models are shown inTable 4 . And the structure and training parameters of the baseline models are detailed in Supplementary Information .