Urban Region Embedding via Multi-View Contrastive Prediction

Authors: Zechen Li, Weiming Huang, Kai Zhao, Min Yang, Yongshun Gong, Meng Chen

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct thorough experiments on two downstream tasks to assess the proposed model, i.e., land use clustering and region popularity prediction. The experimental results demonstrate that our model outperforms state-of-the-art baseline methods significantly in urban region representation learning.
Researcher Affiliation Academia Zechen Li1, Weiming Huang2, Kai Zhao3, Min Yang1 , Yongshun Gong1, Meng Chen1,4 * 1 School of Software, Shandong University 2 School of Computer Science and Engineering, Nanyang Technological University 3 Robinson College of Business, Georgia State University 4 Key Laboratory of Urban Land Resources Monitoring and Simulation, Ministry of Natural Resources
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Data and source code are available at https://github.com/lizc-sdu/Re CP.
Open Datasets Yes We collect a diverse set of real-world data from NYC Open Data1 and use the Manhattan borough as the study area. We utilize the NYC check-in and POI data provided by Yang et al. (2014) for our model training and the popularity prediction task.
Dataset Splits Yes The evaluation results including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R2) are obtained by 5-fold cross-validation, as reported in Table 2.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes In our experiments, the dimension of region representations is set to 96. In the intra-view reconstruction module, we set the number of layers at 3 and the hidden size at 128 for the encoder E(v) and decoder D(v); in the intra-view contrastive learning module, following the settings in Zhang, Long, and Cong (2022), we set the number of positive samples for region attribute and human mobility data at 3 and 4, and the parameter µ controlling the balance between different views at 0.0001. In the inter-view dual prediction module, we set the number of layers at 3 and the hidden size at 96 for F (a) and F (m); in the inter-view contrastive learning module, we set the parameter α at 9. We set the hyper-parameters λ1 and λ2 in the final objective loss at 1. Note that the optimal model parameters are selected using grid search with a small but adaptive step size. To optimize our model, we adopt Adam and initialize the learning rate at 0.01 with a linear decay.