基于用户动态兴趣的单机游戏消费意愿预测

打开文本图片集

中图分类号：TP391.9 文献标识码：A

Abstract： To address the challenges of dynamic user interest evolution and data sparsity in predicting the consumption intentions of single-player game users， this paper proposes a dynamic interest modeling with integrated temporal decay and attention mechanism （DIM-ITDAM） that integrates time-decay weighting and attention mechanisms to enhance prediction accuracy. Firstly， a time-decay function is applied to historical behavioral data，such as game genre preferences and weekly average active duration，to capture the dynamic changes in user interests. Secondly，an adaptive attention mechanism is introduced to quantify the weights of consumption-influencing factors，such as price sensitivity and genre preference，thereby highlighting the role of key factors. Finaly，a fully connected layer outputs the user’s payment probability，constructing an end-to end consumption intention prediction model. To validate the model's effectiveness，experiments were conducted on a dataset comprising 1，2O5 valid questionnaires. The results show that the DIM-ITDAM model achieved a 16.0% increase in AUC value and an 18.9% improvement in F1- score compared to traditional logistic regression （LR）. It also outperformed other comparison models such as Random Forest （RF） and DeepFM. This paper provides an interpretable dynamic prediction tool for the single-player game market， effectively overcoming the limitations of static modeling and collaborative filtering methods in predicting user consumption intentions.

Keywords： consumption willingness prediction; dynamic interest modeling； time-decay weights; attention mechanisms；single-player games

近年来，国产单机游戏市场规模持续攀升，年均增长率突破 15% ，但与该增长态势形成鲜明对比的是，用户付费率仅为 30% 左右，远低于国际平均水平。（剩余8630字）