Research On User Behavior Prediction Integrating Group Behavior And Design And Implementation Of Visualization Platform

With the increasing scale of social networks,user behavior prediction algorithms,which can analyze the dynamic changes of friends among users,extract users' social interest and action time sequence,have gradually become a hot topic in the field of social network research.How to make accurate,reasonable and scientific prediction of users' social behavior on social platform quickly and efficiently has become an urgent demand in many application fields.However,most of the existing user behavior prediction algorithms predict from the perspective of user interest,historical behavior,static network and so on,less considering that the social groups of individuals will also affect the social behavior of users,or it is less considered that users will be affected by the explicit social relationship directly connected with them and the implicit social relationship not directly connected with them when they take actions,and it is less considered in the sequence recommendation or user behavior prediction.When modeling group behavior,the traditional methods such as average strategy or minimum pain strategy also have shortcomings.They can't reasonably gather the relevant information of each user in the group according to the influence of different users,and the evaluation of group behavior modeling lacks a unified standard.In view of the above shortcomings,based on the characteristics of social networks,this paper combines the related technologies of group behavior and behavior prediction,and focuses on how to predict the user behavior of social networks through group division,action embedding and other methods.The main work of this paper includes the following parts.(1)A user behavior prediction algorithm,Gact,which integrates group behavior is proposedThe user behavior prediction algorithm based on group behavior combines user behavior representation with group behavior.The explicit social relationship directly connected with the user and the implicit social relationship not directly connected with the user are considered when making behavior prediction.In other words,through the user behavior representation based on action embedding,the user's own behavior can be used to represent the user,which is more conducive to the subsequent prediction of user behavior.In the representation of group behavior,different weights can be set for different users in the group through the use of attention mechanism,so as to take into account the social influence of different users,so as to overcome the problems existing in traditional algorithms when calculating group behavior or group preference.By constructing the user behavior prediction model based on GRU recurrent neural network,the prediction is feasible and accurate.At the same time,user behavior prediction is the downstream task of group behavior modeling,and its prediction accuracy can be used as the measurement standard of group behavior modeling effect.Through the comparative experiments with other algorithms,it is verified that the Gact model can effectively combine the user's previous actions and the group's actions for behavior prediction,and has better performance than the user behavior prediction algorithm which only considers the user's historical behavior and neighbor behavior.(2)The visualization platform of user behavior prediction based on Gact algorithm is designed and implementedIn order to display and analyze the user behavior prediction algorithm,experimental data sets,group partition results,parameter analysis in the process of experiment,algorithm comparison more intuitively and conveniently,this paper designs and implements a visualization platform of user behavior prediction algorithm based on Gact algorithm,which includes user relationship graph,group partition results,group partition results,User action analysis,algorithm comparison,parameter analysis,data management and other modules,the design of functional modules comes from the experimental process of Gact algorithm.In the visualization platform,user relationship diagram and user action analysis can be used for statistical analysis and query display of experimental data sets;The result of group partition can show the result of community partition,which is the intermediate process of the algorithm,and can select the ideal result of group partition;The interface of algorithm comparison and parameter analysis can compare the experimental results between algorithms and analyze the related parameters.The visualization platform is based on the reality of scientific research,which can efficiently process and intuitively display the experimental data of user behavior prediction,and provide convenience for related scientific research work.