“Paradigm” theory is an important ideological and practical tool for scientific research. The research means and methods of Geographic Information Science follow the laws of four paradigms. Automatic cartographic generalization is not only the key link of map making, but also a recognized difficult and hot issue. Based on large-scale map data and deep learning technology, an automatic cartographic generalization problem-solving model is proposed in this paper. According to the key and difficult problems faced by residential area selection and simplification, residential area selection models and simplification models based on big data and deep learning are constructed respectively, which provides new ideas and schemes to solve the key and difficult problems of residential area selection and simplification.