1. The most frequent relations were identified by counting their occurrences in the training dataset.
2. Four sentences containing these frequent relations were randomly selected from the training dataset.
3. An additional sentence without any relations, randomly selected from the training dataset, was included to add variety.
