Research On User Feature Mining Technology Based On Location Data Of Mobile Terminal

With the development of mobile Internet technology and the popularization of smart mobile terminals,location-based services have attracted more and more attention and have been widely used in civilian fields such as smart cities.As a new direction for the development of cyberspace and electronic countermeasures,the user feature mining technology can extract user features and infer user behavior patterns based on the massive location data,which can support the army and other national power to conduct diversified anti-terrorism tasks for safeguarding national sovereignty and maintaining social security.As a result,research on location-based services has great significance and value.This paper expounds the research background and present situation about user features mining technology,analyzes related theoretical knowledge involved in this kind of technology.Then,based on the previous exposition and analysis,this paper clarifies location data based on smart mobile terminals as the research target,and applies certain scenario to explore user interest regions identification and future regions prediction,which are the parts of user features mining and user behavior pattern inference.Firstly,aiming at the problems of inferior data quality and large data volume in the location of smart mobile terminals,this paper explores the location data preprocessing technology based on the statistical theory.A part of the significant noise points is removed by the speed pruning method,then a trajectory division algorithm based on the angel offset and the distance offset is proposed.By setting the offset threshold,the location points with great change of user behavior are extracted and defined as feature points,which lays the foundation for user interest regions identification and future regions prediction in subsequent chapters.Experiments show that the algorithm has good performance on feature points extraction and operational efficiency.Secondly,aiming at the problem of error clustering caused by uneven distribution density of location data,an improved density peak clustering algorithm based on machine learning is proposed to achieve interest regions identification.The improved algorithm introduces the K-nearest neighbor idea,and the local density is redefined b y the product of the inverse function and the Gaussian kernel function.The critical point of the cluster center is determined according to the slope change trend of the weights in the sorted graph,and the automatic selection of the cluster center is realized.Then,combined with Google reverse geocoding,the latitude and longitude coordinates of extracted regions of interest are converted into corresponding local semantic names to realize the transition from the data layer to the semantic layer.Experiments show that the improved algorithm can accurately identify the user's regions of interest.At last,aiming at the poor prediction accuracy of low-order Markov model and the high sparse rate of high-order Markov model,a multi-order fusion Markov model based on Adaboost algorithm is proposed.The model firstly determines the model order k by the matching of the prefix trajectory sequence and the historical trajectory sequence,and uses the Adaboost algorithm to assign the corresponding weight coefficient according to the importance degree of the 1?k-order model.Finally,a multi-order fusion Markov is generated.The model is used to predict the regions that the target user ma y access in the future.The experimental results on the real user trajectory dataset show that the Adaboost-Markov model has good predictive performance and universality.