Design And Implementation Of A Mobile Game User Loss Prediction System

With the diversification of entertainment diversity and the competition in the mobile game industry,the loss of players has become a problem that the game business layer needs to solve.Effective retention methods and reflow strategies can improve the overall profit of the game.However,the existing scheme is limited to specific games,and the log data is static data rather than dynamic data.The existing scheme only targets the player churn prediction without further guessing and analysis of the cause of the churn.In this thesis,we mainly design the data collection,data aggregation and loss prediction analysis from the relevant big data technology in the open source community,and use the long-term and shortterm memory network(LSTM)to improve the prediction by using the player's time series data.Accuracy,and using statistical results to further predict the cause of loss using statistical analysis.This thesis focuses on three issues of reasonable storage of player data,real-time analysis and aggregation,and offline prediction loss analysis.The main work of this thesis includes: Firstly,through the preliminary research and analysis of the demand of mobile game user churn prediction system,complete the outline design;secondly,realize the system with big data technology and deep learning technology: collect log data with Flume,Kafka as log data buffer Middleware,according to the characteristics of the game log data,select the HBase database to store data,and optimize the HBase according to the requirements,select Flink as the computing platform according to the demand scenario;Finally,predict the lost players for a certain game,select the more common statistics according to the survey.Indicators,using Keras to train the loss prediction model,and using Fens to use the Tensor Flow Java API to call the training model for prediction,effectively reducing the length of the prediction results,and finally sort out the games that cause the player to lose in each type of game event in different versions of the same game element.This thesis designs and implements a mobile game user churn prediction system,which can realize log collection,log aggregation and offline prediction.The test and verification show that the system meets the forecasting requirements.This thesis has reference value for the realization of the mobile game user churn prediction system,the game business layer to provide predictive churn player list and cause analysis report,and the data support for the game business layer retaining players.