As new social media platforms including games, self-media, and online forums have emerged, people’s daily social interaction strategies have also evolved. This study suggests a technique to improve game suggestion and operation by properly predicting players’ preferences by combining the social ties and spatiotemporal activity of game users. This model extracts the social interaction network between players and time-series behavior data from public game logs. It uses the Graph Sample and Aggregate (GraphSAGE) model to learn the social features between players, and employs Long Short-Term Memory (LSTM) to capture the temporal dependencies of different users. The model also incorporates a Self-Attention Mechanism (SAM) to weighted fuse the two types of features, and finally realizes interest prediction. The verification results on the publicly accessible dataset, Mobile Game User Data Analysis dataset, (The link is: https://www.kaggle.com/code/edinaschmidt/mobile-game-user-data-analysis ) show that the proposed model performs best on the test set. The Accuracy reaches 0.901, Precision is 0.887, Recall is 0.872, and F1-score is 0.879. Compared with models that use only temporal behavior or social features independently, the F1-score is increased by an average of about 3% points. Compared with traditional fusion methods, each indicator also has obvious advantages. This indicates that the feature fusion method combining social relationships and temporal behavior can predict players’ interests more accurately. It also gives people a more intuitive understanding of how players’ social interactions and behavior patterns interact and affect interest preferences. Therefore, game developers should take this multi-dimensional interaction relationship into more consideration when designing game recommendation and operation strategies. They should improve players’ experience from the user perspective and enhance the stickiness between games and users.
Wang et al. (Tue,) studied this question.