Research And Application On University User Similarity Based On Trajectory Data

The application of clustering and recommendation system needs the parameter of user similarity.At present,there are many methods to measure user similarity.The method based on trajectory data is one of the important methods.With the increasing popularity of location acquisition technology,such as GPS and Wi-Fi,a large number of user trajectory data are acquired,which makes this method become a hot research topic.At present,most of the trajectory-based user similarity studies are mainly applied in the field of community network,and rarely reported in the physical world.Campus network is a good platform for researching the user behavior mode of WLAN because of its penetration and diversity of users.Under the above background,based on user trajectory data collected from University scenes,two user similarity measurement methods are proposed for clustering and recommendation system application.The main work of this paper is as follows:The research status of user similarity based on trajectory is analyzed and summarized.Common methods of trajectory similarity measurement,clustering technology and recommendation system technology are introduced.Introduce the trajectory data collection based on real university scenes,and give the steps of trajectory data preprocessing,including data cleaning,data integration and data conversion.In addition,multi-dimensional data mining provides data support for subsequent research.User similarity research based on L2 norm normalized Euclid distance(2NN-Euclid distance).The feature vectors are constructed by using the location and duration of the user's presence.Then,the 2NN-Euclid distance is proposed and the user similarity is studied based on 2NN-Euclid distance.User clustering results are used to evaluate the quality of similarity measure.The experimental results show that the clustering result based on 2NN-Euclid distance is better than that based on traditional Euclid distance,and it has better robustness to noise.User similarity research based on improved HGSM method.Using a hierarchical graph-based similarity measure(HGSM)framework,the location history of human is modeled,and the similarity between university users in geographic space is explored in order and hierarchy.An improved method of HGSM based on search step size threshold is proposed.The experimental results show that the improved scheme improves the computational efficiency while guaranteeing the validity of the recommendation list compared with the traditional HGSM scheme.