The goal of sequential recommendation is to gain valuable insights from previous interactions between users and items to predict the next item that the user maybe interest. In this research, an enhancement of the Meta Transitional Learning (MetaTL) framework is introduced, known as the Category-Aware Transitional Meta Learner (CAT-ML). The CAT-ML model combines a category-level transition meta-learner and an item-level transition meta-learner. By utilizing the category-level transition meta-learner, the proposed model effectively captures user behavior patterns by initially acquiring general features from behaviors at the category level. Subsequently, the feature representation obtained from category transitions is inputted into the item-level transition meta-learner, where an attention mechanism is employed to guide the extraction of behavior features from interactions at the item level. The experiments conducted on the Foursquare Check-in Dataset demonstrate that the CAT-ML model outperforms the MetaTL model, exhibiting improvements of 10.2% in top one item hit rate and 23.8% in category hit rate. Notably, the CAT-ML model demonstrates superior performance in scenarios involving cold-start users or new items in user history behavior, surpassing the MetaTL model by a significant margin.