Research On Personalized Recommendation Based On The Data Of Mobile Crowdsensing

In recent years,the rapid popularization and application of mobile smart devices has promoted the rapid development of mobile crowdsensing.Users who participate in mobile crowdsensing provide a large amount of location data on the Internet crowdsensing plat-form,which could be used in the area of intelligent transportation system,social recom-mendation and urban dynamic sensing.These crowdsensing location data are connected in time sequence order and form rich trajectories.It is feasible for us to obtain users’pref-erences,recommend personalized trips for users,and investigate the changes of pedestrian flow by using the large amount trajectories of users.However,the study of personalized recommendation suffers from the following challenges:（1）The traditional personalized trip recommendation mainly focuses on the popularity of the Point of Interest,and the popularity of routes between them is not fully used,which may also bring extra user ex-perience.The trade-off between POIs and routes in the personalized trip recommendation needs further studied.（2）In the traditional prediction of pedestrian flow,the flow brought by the attraction of events on the routes is not fully explored and needs further studied.（3）In the traditional data sensing scenario,the data source generates the sensing data and transmits them to the cloud.Because the generation rate of crowdsensing data has an im-pact on the quality of sensing data,the data generation rate of crowdsensing needs to be further studied.By taking the advantage of the rich crowdsensing location data,we integrate the lo-cation data of crowdsensing and study the personalized trip recommendation with POIs and routes,the impact of business events on pedestrian flow prediction,and the problem of personalized business events placement.The rich crowdsensing location data can be uploaded through infrastructure networks（such as 4G,5G,Wi Fi,etc.）.They can also be transmitted with the help of emerging networks such as the Internet of Vehicles.The data source uploads the sensed data to the server,and only relying on the infrastructure network cannot meet the transmission requirements of various crowd-sensing data.There-fore,this dissertation starts with the vehicular crowdsensing scenario and investigates this new transmission method.The age of information as a metric describes the data trans-mission performance in mobile crowdsensing to meet the requirements of data quality in crowdsensing.The main contributions of this dissertation are as follows:（1）To solve the problem that traditional personalized trip recommendation mainly focuses on the popularity of the Point of Interest,but ignores the popularity of routes be-tween them,a personalized trip recommendation algorithm that not only considers Points of Interest but also consider the attractive routes between them is proposed,so as to en-hance the user experience of a recommended trip.Based on the popularity of Points of Interest and their Gini coefficient,we discover the attractive routes between Points of In-terest from the dataset.Then,we adopt the gravity model to estimate the rating score of the attractive routes in the category space.Furthermore,we proposed a personalized trip recommendation algorithm combining both Points of Interest and attractive routes under multi-constraints.To the best of our knowledge,we are the first to introduce attractive routes to personalized trip recommendation.Experimental results show that the perfor-mance of our proposed algorithm outperforms the baseline algorithms in terms of recall,precision,F₁score,and total user experience.The main reason is that the user experience is improved by incorporating attractive routes that users prefer in the recommendation process.（2）To solve the problem that the pedestrian flow between points of interest changes due to the influence of business events placed the routes between them,an attraction-based matrix factorization model is proposed to predict the pedestrian flow with different busi-ness events efficiently.A pedestrian flow matrix is constructed by extracting the user’s historical trajectories from the sensing dataset,where each element is the pedestrian flow size between two Points of Interest.With the idea that different categories have different attractions to pedestrian flows,we propose an attraction-based matrix factorization model.After the parameters estimation and optimization of the model,the pedestrian flow pre-diction is made according to the optimized parameters and prediction model.This method can predict the pedestrian flow between Points of Interest if the set and position of business events are determined.Experimental results show that the proposed model is superior to other probability-based matrix factorization models in terms of Root Mean Square Error（RMSE）and matrix similarity.（3）It is often difficult for operators to decide which routes are fit for holding business events to attract more pedestrian flows,which is a derivational problem from pedestrian flow prediction with business events.A personalized route recommendation model for business events is proposed to solve this problem.Based on the users’location data,we first extract users’trajectories based on their location data.Then we leverage the Skip-gram mode to learn the latent representations of routes.In addition,based on the historical position records of business events,we improve pair-wise ranking loss to a flow-aware-based method to learn events’latent representations.Finally,according to the ranked point product of learned latent vectors of routes and an event,the top-N routes are recommended for an event,which can help operators make decisions.The experimental results show that the proposed method is superior to other baseline algorithms in terms of accuracy and Mean Reciprocal Rank（MRR）because the learned latent vector representation considers the context information and incorporates flow factors.（4）To solve the problem that the traffic hole affects the data transmission in the vehicular crowdsensing network and therefore affect the data quality,an algorithm of the optimal generation rate at the source by considering the trade-off between the AoI（Age of Information）and transmission cost is proposed.Based on the established data delivery model with AoI,we aim to optimize the sensing data generation rate to minimize the total average cost of AoI and transmission.This method can achieve the lowest total average cost by using the optimal generation rate to generate sensing data,resulting in lower AoI and transmission costs.Experimental results show that in terms of the total average cost,our proposed algorithm is superior to the randomly generated sensing data algorithm and the algorithm of generating sensing data at will.