Study On User Group Profile Based On Wireless Sensing

With the rapid development of various Internet technology and applications,the technology of user profile has become a hotspot in academia and industry,and has been widely used.At the same time,the construction and development of wireless network in colleges are also gradually improved,resulting in a large number of user access data.Research on user group profile based on wireless perception is inevitable and has very important theoretical and practical significance.In the research of user group profile in wireless networks,user's location is the most important feature.Therefore,this paper firstly proposes a POI(Point of Interest)positioning algorithm based on PU(Positive-Unlabeled)learning and generative adversarial networks to acquire user's accurate location.The concepts of generator and discriminator are introduced into positive and unlabeled samples,which solves the problem of insufficient positive samples in data set,improving the accuracy and stability of location,and the proposed method has universal applicability.Secondly,this paper uses cosine similarity to measure the temporal and spatial similarity between users,and combines parameter-adaptive DBSCAN clustering algorithm to cluster users,so that we can find user groups.Finally,tag informatization is carried out in each user group from four aspects: basic attributes,user grading,behavior preferences and POI prediction.A complete user group profile is constructed according to the idea from individual to whole,and is presented in the form of front pages.POI prediction label is most important and developmental.In order to predict the future POI that users will sign in accurately,an algorithm of collaborative filtering POI prediction based on time weights in the clusters is proposed in this paper.Considering the rule of human activities and the influence of time on the prediction results,this algorithm introduces time weights into the user's POI score matrix.It makes the user's latest activities contribute more to the prediction results than that in the early stage,so the approach has better accuracy and recall rate.Several groups of comparison and analysis show that POI location algorithm based on PU learning and generative adversarial networks as well as POI prediction algorithm based on clustering and time weights have better accuracy and stability than traditional methods,and provide reliable location features and basis for drawing user group portrait.In addition,the study of user group profile based on wireless sensing outlines the characteristics,behavior patterns and rules of users from a macro perspective,which provides important reference value for campus managers and builders of campus wireless,and also provides guiding significance for the study of user profile in other fields.