Research on RPG Game Recommendation System

Abstract

In RPG Games, how to make players quickly adapt to the game and like the game has always been the focus of game merchants. Good props recommendation can not only improve the game experience of players, but also bring benefits to game providers. RPG games are different from other types of games in that the player level changes faster, and there is a great correlation between levels and equipment. Therefore, we need to recommend corresponding props in real time according to player’s levels. In the previous recommendation systems, offline features are often used, which leads to the inability to capture the level changes of players in real time, and finally affects the recommendation results; Similarly, in the process of offline sample splicing, there will be inconsistency between the characteristics used in online services and those used in model training samples. To solve the above problems, we use real-time computing to replace the previous offline framework. By real-time changing features, we avoid the problem of outdated features, and use real-time sample splicing instead of offline splicing to directly save the samples requested and purchased by users to HBase, to solve the problem of inconsistency between online and offline features. The experimental results show that after using the above scheme, the real-time recommended exposure ARPU effect is 14.75% higher than the offline effect. Finally, by solving the problem of data skew, the overall performance of the system is improved by more than 50%. The experimental results prove that the model can well capture the player’s status changes after the features are real-time, and the use of real-time technology to ensure the consistency of online and offline features significantly improves the prediction effect of the model.