Identifying Urban Functional Zones By Considering Multi-Spatial Distribution Patterns Of Points-of-Interest

Urban Functional Zone(UFZ)can reflect the layout of urban spatial structure and carry social and economic activities.Identifying UFZ is crucial for understanding the complex urban system,scientific resource allocation,urban planning and management,and further promoting urban sustainable development.Points of Interest(POI)data have been widely used to identify UFZs due to their rich semantic information of urban function.POI data is transformed into geo-corpus by employing specific sampling strategies.Natural Language Processing(NLP)techniques can be used to extract functional semantic features,which help identify UFZs with high accuracy.However,the current POI sampling strategies in the construction of geo-corpus ignore the spatial interaction which has a significant impact on the functional semantics of POIs.Additionally,previous studies only focus on a single spatial distribution pattern,which is difficult to comprehensively and accurately reveal the semantic meaning of urban function.Moreover,the difference in identification performance of UFZs based on different spatial distribution patterns has not been studied.To address these research gaps,this study proposed a UFZ identification method by capturing POI multi-spatial distribution pattern from the perspective of synthesizing multispatial distribution pattern information to describe urban functional semantics.First,a POI spatially-explicit sampling strategy was designed to capture the corresponding spatial distribution pattern,taking POI spatial interaction into account.Meanwhile,three existing POI sampling strategies were introduced to construct geo-corpuses that capturing different POI spatial distribution patterns,and the Word2 Vec model was trained to extract POI embeddings corresponding to each spatial distribution pattern.Then,feature fusion was used to enhance the spatial distribution information of POI embeddings,and the UFZ embeddings were calculated to represent the functional semantics.Random forest model was leveraged to classify all UFZs.Finally,several groups of comparative experiments were designed to validate the effectiveness of the proposed method.The experimental results show that the proposed method can effectively identify UFZs(OA=0.729,Kappa=0.618),significantly outperforming the baseline method that considers single spatial distribution pattern(ΔOA=0.085,ΔKappa=0.122).This study measures the contribution of different POI spatial distribution patterns to the identification of UFZs.In addition,the results show that the spatial distribution patterns have a representational tendency in functional semantics.The proposed method provides a more accurate method for identifying UFZs,which can help planners understand the functional semantic representation of UFZs from the perspective of POI multi-spatial distribution patterns,and provide references for POI-based urban-related research.