Research On Context Awareness Based Multi-modal Transportation Recommendation Method

With the rapid development of public transportation,the available travel options are becoming more and more diversified,and people’s requirements for the travel mode are getting higher and higher.Travel demand has gradually been developed from the previous single mode(such as bus,taxi,bicycle)to the current single or multi-modal combination(such as the combined transportation of bus and taxi,the combined transportation of bus,subway and bicycle).Therefore,in the face of diverse travel modes and complex transportation system,how to provide travel plans that meet user preferences at the right time and at the right place has become one of the hot issues in the field of smart transportation.In response to the above-mentioned problems,this paper mainly conducts research from the following two aspects:(1)In order to solve the problems of considering only one transportation mode and neglecting user preference in the regional transportation recommendation problem,and class imbalance problem in multi-class task,a regional multi-modal transportation recommendation method based on Particle Swarm Optimization and Light GBM was proposed.This method can comprehensively consider the user’s travel preferences in terms of time,space and travel cost,and uses mathematical statistics and representation learning methods to capture the internal relationship between user travel and various elements.At the same time,in order to alleviate the negative impact caused by the imbalance of sample class,the index optimization method based on particle swarm optimization algorithm is used to search for the optimal weight for each class,and the prediction results of the model are modified to achieve the purpose of maximizing the evaluation index.Experimental results show that compared with traditional algorithms,the model proposed in this paper has better performance in spatio-temporal feature extraction,alleviating class imbalance and recommendation accuracy.(2)In order to provide users traveling in multiple regions with a travel mode that not only meets their travel preferences,but also conforms to the current regional traffic,a multi-region and multi-modal transportation recommendation method based on graph embedding and Ca GBDT is proposed.This method uses the graph embedding representation learning method and the Compressed Interaction Network to automatically extract the travel rules of users in different regions and in different temporal-spatial contexts in the recommended scenarios of Top1.At the same time,in order to realize the deep-level representation learning of features and alleviate the negative effects brought by the imbalance of categories,the Ca GBDT model is constructed for high-performance classification based on the ideas of cascade structure and residual learning.Experimental results show that the proposed model can better mine user preferences in multi-region and multi-modal transportation recommendation task,and has better recommendation accuracy and stability.