Research On User Active Value Prediction Based On Forum User’s Time And Behavior Data

With the continuous development of mobile networks,social networks have become an important platform for network users to exchange and disseminate valuable information.More and more netizens have begun to use online forums for various information exchange activities.Along with various behaviors of users in social networks,a large amount of user historical data is generated and accumulated.The user’s historical data contains the user’s behavior characteristics and the time characteristics associated with the behavior.Predicting the activity value of users in social networks is beneficial to retaining and maintaining users,and has a profound significance for the operation of social networks.In this context,based on the analysis of various behavioral data and time data of users in the online forum,this paper builds a prediction model of user activity based on the time and behavioral data of forum users.First,for the problem that the user’s behavior data in the experimental data is scattered in different documents and there are missing values,this paper integrates the data and fills in the missing values during the data preprocessing process;for the labeled user active values in the experimental samples For the problem of small sample size,this paper proposes the COREG-GASVM algorithm.The COREG-GASVM algorithm first uses genetic algorithm to optimize the SVM algorithm,and then improves the semi-supervised algorithm COREG based on the genetic algorithm optimized SVM algorithm to expand the number of experimental samples.Secondly,by analyzing the historical data of users in social networks,this paper proposes a combination method based on behavioral characteristics and temporal characteristics,and screens the behavioral characteristics and temporal characteristics for the problems of complex and diverse historical data and unclear effective characteristics.The screening method based on behavior characteristics is mainly to use gradient lifting trees to filter out the features that meet the requirements;the screening method based on time characteristics is mainly to divide the time by month,and experiment with the "last landing time" as the influencing factor.Thirdly,a user active value prediction model based on Stacking integrated learning is proposed,and the Stacking algorithm in integrated learning is used to predict user active values.The model uses support vector machines and K nearest neighbors as the primary learners,and uses gradient boosting trees as the secondary learners.Among them,the training features of the primary learners are the features after screening the behavioral features and temporal features.The training features are all features,and finally the user’s active value is predicted.Finally,through the use of user behavior data in a computer professional field for experiments,the effectiveness of the improved COREG-GASVM algorithm,feature selection,and user activity prediction model based on Stacking integrated learning is verified.