Study On Railroad Passengers’ User Profiles And Scheme Recommendation Based On Multi-source Data

With the construction of infrastructure such as high-speed railways,people have more and more options for train travel.How to quickly choose the one that suits you is more and more important,especially in large cities where there are hundreds of trains,and the ticket purchase interface can only Manual screening by simple sorting;at the same time,based on the rise of the concept of MAAS,people are increasingly demanding the improvement of travel services and it is of great significance to efficiently provide passengers with suitable railway travel solutions;on the other hand,the recommendation system and user profiles system are widely used in the Internet field such as e-commerce marketing,which has greatly improved user online experience.This article first discusses the current status of user profiles,inter-city travel plan selection,and recommendation systems,and determines related technologies for establishing railroad passenger profiles and train recommendations.Through the cleaning,expansion and integration of Ctrip user train travel history records and online crawling related information from multisource,using statistics,KMeans clustering,Xgboost and other algorithms to construct three levels(data layer,label layer and application layer),four types of labels(User attributes,behaviors,preferences,sensitivity)railway travel passenger user profiles system.It has been experimentally verified that the sensitive labels F1-Measure(Score)based on Xgboost modeling are all above 90%,with high accuracy.Use micro-word cloud,ECharts and other software to realize the personal level and group level visualization of user profiles.Combining the railroad passenger user profiles system,relying on the user space vector(Vector Space Model)model,cosine similarity and other methods to construct railway trip recommendations.Taking Beijing-> Tianjin-Beijing-> Tianjin,Beijing->Shijiazhuang,Shijiazhuang-> Tianjin 3 pairs of train lines as an example,in terms of train number recommendation,the algorithm in this paper is compared with the current widely used time-spent-based and time-sequentially-based methods It shows that the method proposed in this paper has higher indicators on the three types of train lines than the other two methods,which can improve the user’s ticket purchasing experience.