Research On Recommendation Of Friends Based On Trajectory Mining And Feature Vector Fusion

In recent years,with the popularity of the running software and so on,friend recommendation algorithm based on trajectory is gradually becoming a hot research.Therefore,how to make use of the user's track data to find out the potential and meaningful trajectory model has became a hot issue,then friend recommendation is developed on the basis of it.This paper firstly introduces the architecture of the sports social platform based on Android,the trajectory acquisition module and data upload module are mainly introduced.Then the recommendation system theory is introduced,and the following research is carried out around the fusion recommendation algorithm based on trajectory mining and user feature.(1)This paper describes the user's hot trail,stay point,track region and correlation algorithm of trajectory mining.Aiming at the problem of traditional DBSCAN algorithm,such as the complexity of regional query and human participation in parameter setting.This paper presents a ?-ADBSCAN algorithm to dig the hot trail,hot trail reflect users frequent movement patterns,which can represent some of the user's behavior habits.(2)The ST-Cluster algorithm is designed based on the hot trail to mine the user's stay point.This algorithm can prevent the loss of the trajectory data caused by the poor GPS signal and the noise caused by the road congestion.Then the Pearson formula is introduced,the time similarity factor and the hierarchy factor are added in order to calculate the stay point similarity between different users.(3)The trajectory segmentation method is defined for the representation of the user's trajectory,then user's trajectory is transformed into the track region consisting of the segmentation points and hot trail.Track region similarity is calculated by the MHTR block sequence which is composed by MBR algorithm.And the overall similarity calculation of trajectory behavior combines above two similarities.Compared with the traditional friend recommendation algorithm based on the stay point,user similarity can be better measured from different scales.(4)Aiming at the problem of cold start and singularity of recommendation,this paper studied the inherent characteristic similarity of users by mining the user's interest matrix through the registration information and building user's social matrix according to the number of friends attenuation,then the inherent feature similarity is calculated based on the modified cosine similarity.Finally,this paper studied the fusion method of trajectory behavior similarity and inherent characteristic similarity for friend recommendation.This part designed a fusion recommendation algorithm based on dynamic threshold,the experimental results show that the fusion method is superior to a single algorithm in performance.