Research On Online Individual Feature Recognition And Behavior Prediction

Through human behavior,Social sciences and humanities reveal the laws behind social and economic phenomena.However,as the network develops,human behavior is becoming more and more complex,and it is becoming difficult to discover the hidden information behind it,the big data technology provides a new solution to this problem.Online user behavior is the manifestation of human behavior on the Internet.Through the collection and analysis of online user behavior data,it can provide more accurate data support to help government and enterprises to make a decision and solve other issues,and can also provide theoretical support for scientific researchers when studying human behavior.In this paper,through the online user's behavior,characteristics and click stream data,research on anonymous user identification and predict users'next click.Analysis of the factors that affect the recognition and prediction accuracy.The main research content includes the following two aspects:Firstly,Anonymous user identification based on multi-dimensional trajectory set.Through experiments,it is verified that the user's software clickstream data contains the user's behavioral rules,and the recognition accuracy is improved compared with the web browsing data.An improved(Anonymous User Identification,AUI)algorithm based on association rules is proposed,which reduces the time cost of the algorithm by filtering abnormal data and reducing the number of scan data sets.At the same time,a(Multidimensional Trajectory Set,MTS)model was proposed,which established a unique multidimensional trajectory set for each user through AUI association algorithm and vectorization method,and combined with maximum entropy natural language processing to identify users.Experiments show that the accuracy of MTS model is 5.09%higher than the supple-based Profiling(SP)algorithm,and 10.6%higher than the C4.5decision tree.Secondly,User click behavior prediction model based on reinforcement learning.Build a basic reward matrix B-Reward through a clickstream network to analyze the user's overall click behavior,and a weighted reward matrix W_i-Reward through frequent item sets to analyze the next click when the user in a specific click chain.Combining two matrices to propose(Combination Matrix Q-Learning,CMQ)prediction algorithm.At the same time,a user click behavior prediction model(Reinforcement Learning-Prediction,RL-P)based on reinforcement learning is proposed,which analyzes the parameters that affect the prediction accuracy and combines the CMQ algorithm to predict the user's next click behavior in the current state.Experiments show that the average prediction accuracy of the RL-P model is 88.28%.As the number of known historical click sequences increases,the accuracy of the prediction model also increases,with an average increase of 2.61%.