Text this: Structural self-contrast learning based on adaptive weighted negative samples for facial expression recognition