Study On The Identification Method Of Urban Functional Zone Integrating POI Local Spatial Relationship And Global Semantic Information

The identification of urban functional zones aims to recognize and classify the socioeconomic functions of different areas within a city,which is crucial for urban planning and management.With the accelerated modernization of cities in our country,the spatial structure of cities has become increasingly complex,making the research on urban functional zone identification a hot topic in the academic community.In recent years,the rapid development of big data technologies has provided abundant data sources for research on urban functional zone identification,including high-resolution remote sensing imagery,Point of Interest(POI)data,and social sensing data(such as mobile positioning data,public transportation card data,social media check-in data,etc.).Among them,POI data is an important data source that contains socioeconomic attribute information and spatial location information of points.Compared to high-resolution remote sensing imagery,POI data has advantages such as easy accessibility,good timeliness,high accuracy,and the ability to reflect the socioeconomic characteristics of urban functional zones.Compared to social sensing data,POI data has advantages such as comprehensive coverage and not involving user privacy.Therefore,this study selects POI data as the data source for urban functional zone identification research.POI represents the geographical entities that constitute urban functional zones,and both its semantic features and local spatial correlation features have important influences on the types of urban functional zones.Existing studies have separately utilized POI semantic features or spatial distribution features to extract urban functional zone characteristics and achieve classification and identification of urban functional zones.However,there has been no research that integrates the two approaches to fully exploit POI information for functional zone identification.Therefore,this study proposes the fusion of POI’s local spatial features and global semantic information to achieve urban functional zone identification.It utilizes the Place2 Vec model to extract the local spatial features of POIs and employs the Latent Dirichlet Allocation(LDA)model to capture the global semantic information of POIs within functional zones,thereby establishing the urban functional zone identification model based on Place2 Vec and LDA(referred to as the Place2Vec-LDA model).Specifically,this method first utilizes the Place2 Vec model to embed POI points into a high-dimensional vector space to extract their local spatial features and obtain vector representations of different POI types.Then,it employs the LDA model to perform topic modeling on the research area and extract the global semantic information of POIs within functional zones,obtaining the probability distributions of topics within functional zones and the probability distributions of POI types within topics.By combining the POI vectors and these two probability distributions,the feature vector of a functional zone is computed,achieving the fusion of POI’s local spatial features and the global semantic information of POIs within functional zones.Finally,the Support Vector Machine(SVM)model is utilized to implement urban functional zone identification.This study takes Chaoyang District in Beijing as an example to experimentally validate the established Place2Vec-LDA model.By comparing the results with the manually identified urban functional zones in the area,the effectiveness and accuracy of the model are demonstrated.Furthermore,a comparison is made between this model,the Place2 Vec model,and the LDA model.The experimental results show that this model achieves higher overall accuracy(74%)and Kappa coefficient(0.67).