Air quality prediction is a crucial issue in air pollution control and plays a vital role in environmental preservation and the promotion of sustainable development. A novel air quality prediction framework, referred to as Correlation-split and Recombination-sort Interaction Networks (CRINet), is proposed in this paper. Firstly, the data is divided into smaller segments based on the Correlation-split strategy, and is followed by a convolution operation to enhance the capability of extracting essential features. Secondly, Recombination-sort strategy is used to facilitate the interaction between temporal features. The CRINet model utilizes a dual-layer CRINet network structure to extract internal dependencies and temporal features from air quality data for prediction. Finally, the Beijing PM2.5 dataset and the Beijing Shunyi-station Air Quality dataset are used for experimental evaluation. The outcomes demonstrate that the CRINet model surpasses other models in prediction accuracy, particularly in the prediction of multiple pollutant concentrations. This provides theoretical and methodological support for accurately assessing air quality and formulating treatment strategies for relevant departments.