Tag
The paper proposes UNIT, a framework that fine-tunes a large language model on the first task to enhance its adaptability for graph continual learning, addressing semantic-structural separation and imbalanced knowledge transfer. Experiments show state-of-the-art performance.