A Business Hall Location Model Based On Semi-supervised Learning Ranking Network

Site Selection is an import task in business.A good strategy for site selection is crucial and import for those who want to start or expand their business.The core problem of site selection is how to reasonably and accurately evaluate the candidate regions in urban areas,which will bring significant impacts and potential for subsequent decision process.Besides,while managing the business stores after the site selection,how to analyze and predict the flow volume of passengers or customers can also provide import help for the activities of services adjustment,sales promotion,commodities selection and so on.However,site selection and flow volume prediction are both difficult tasks.Due to the fact that they are influenced by multiple complex factors,such as the geographical positions,road network structures,transportation convenience,POIs distribution,demographical features,economic consumption features and so on in the region.Meanwhile,different regions far and near also have complex correlations and have impacts on each other.For example,regions with similar functions tend to have similar business potential.As for flow volume prediction,it is dynamic and change with time,for each region,its flow volume is not only influenced by the trend of its history flow,but also correlated with other regions,as well as some external features,such as weather,temperature,holidays,and so on.For the sake of multiple influencing factors,traditional methods for site selection are mostly based on the domain knowledge of human experts,they design some indicators and features to obtain a rule to rate the candidate regions,and finally choose some reasonable regions to build the stores according to the ratings.There is no doubt that it is time consuming and painful in the traditional way also with no effective manner to evaluate the results.And for flow volume prediction,there only exists some techniques to help them collect some historical data and make some analysis and how to accurately prediction the flow volume in the future is needed.To this end,in our paper,we attempt to analyze and model on these complex factors to deal with the site selection task for business halls and flow volume prediction task using deep learning techniques.We propose a semi-supervised ranking networks for site selection task and a multi-view graph convolution network model for flow volume prediction.With our model,we can help investors to make better site selections decisions and moreover help them make accurate prediction of flow volume in their stores,which can help to better serve more people as well as efficiently manage the their business.More specifically,we firstly divide the target city into regular or irregular regions,then our site selection task and flow volume task will be based on these regions.Now,some existing business halls will be located in some regions,these regions can collect total revenues of corresponding business halls and use them as ranking labels to guide our model learning.At the same time,we collect the transition flow of humans and build a human mobility graph to model correlations of different regions via attribute network embedding.In each region,we also collect various features from different sources.Our ranking model will learn to rank candidate region in the city and provide the import information for subsequent site selection decision.And for flow volume prediction,we take into consideration the multi-view historical flow features and the correlations between different regions far and near as well as external features,our model learns from cross-domain features and achieve an efficient and accurate results on flow volume prediction task and provide more valuable information for the management of business halls.