Joint Gated Co-attention Based Multi-modal Networks For Subregion House Price Prediction

Urban housing price prediction is widely accepted as an economic indicator which is of both business and research interest in urban computing.And it will greatly help the planning and construction of smart cities.However,to predict the variation of ur-ban subregion housing price,there are existing many challenges including the complex influencing factors,the sparse property of transaction records,and the strong spatial-temporal correlations.Hence,it is challenging to implement such a model.Firstly,to address these challenges,in this work,we study an effective and fine-grained deep spatial-temporal model for urban subregion housing price predictions.Compared to existing works,our proposal improves the forecasting granularity from city-level to mile-level with only publicly released transaction data,which overcomes those challenges and realizes the accurate prediction.Next,we employ a feature se-lection mechanism to select more relevant features.Our model can be divided into four submodules including(long-term and short-term DenseNet,current ingredients,and Kalman Filter based future expectation module)to capture the correlations of long-term,short-term,current ingredients and future expectation among subregions.Based on DenseNet,this article adopts a dense connectivity connection method to learn the characteristics of various dimensions,so as to better reduce the impact of data sparse-ness.In the current ingredients module,our work specifically adopts the feature selec-tion strategy to filter the influencing factors,so as to further avoid overfitting caused by information redundancy.Besides,we devise a novel JGC based fusion method to better fuse the heterogeneous data of multi-stage models by considering their relationships,because various stage models interact with each other in temporal dimension.Then,we propose an integrated model,JGC_MMN(Joint Gated Co-attention Based Multi-modal Network),to learn all-level features and capture spatiotemporal correlations in all-time stages.Further,our JGC can better capture the correlations between heterogeneous submodules than other related works..Finally,we select the real-world house price transaction datasets of Beijing and NYC as the empirical data for our model.Extensive empirical works compared to related models including SVR,VAR ST-ANN,Deep-ST,ST-InceptionV4,ST-ResNet show that our model can improve the value of RMSE by 23.1%,27.2%,21.8%,21.1%,19.2%,and 17.8%,which demonstrates the effectiveness of our proposal,and this fine-grained housing price forecasting has the potential to support a broad scope of applications,ranging from urban planning to housing market recommendations.Be-sides,our model considering the all-time stage design as well as the JGC based fusion can be applied to other similar domains in urban computing.