Accurately identifying essential Points of Interest (POIs), modeling resident-POI dependencies, and quantifying spatial structures are fundamental to creating livable environments. We propose a representation learning framework for quality-aware modeling of community life circle structures. Our method integrates urban mobility patterns to resolve two key limitations of existing approaches: rigid spatial boundaries and ambiguous dependency relationships. First, we employ k-nearest neighbors (kNN) to adaptively identify candidate POIs, eliminating fixed-range constraints. Second, we design a POI quality index using multi-source urban data and compute community-POI dependency indices through synergistic integration of quality metrics and category weights. Third, temporal segmentation via KL-divergence constructs activity graphs reflecting time-variant resident mobility patterns. Finally, an autoencoder-based representation learning module embeds these graphs to generate structural feature vectors. For the convenience prediction application, experimental evaluations were conducted using real datasets to verify the effectiveness of the proposed representation method for community life circle structure.