王者荣耀游戏实时胜率预测

Real-time winning rate predication of Honor of Kings

王者荣耀游戏是当前最受欢迎的手机游戏之一。为了增强玩家和观众的游戏体验,越来越多的游戏公司研发人工智能(AI)解说员用来同步解说游戏实况,而游戏的实时胜率预测和关键数据特征是同步解说的重要内容。现有的关于王者荣耀胜率预测的研究主要集中在历史游戏数据驱动的技术上,因此,预测率和可解释性都不理想。利用王者荣耀游戏的真实对战数据,结合角色之间的交互作用,提出了一种新的角色-实时联合网络(RRSN)。最后,通过大量真实数据实验对王者荣耀实时胜率进行预测,准确率能达到87%,模型在预测精度和可解释性方面比其他的模型效果更佳。利用本文提出的预测模型和算法,AI解说员可以在解说词中给出较为准确的胜率预测与分析,大大增强游戏观看者的体验。

Honor of Kings(HK)is one of the most popular mobile games currently.In order to enhance the game experience of players and audiences,more and more game companies are developing artificial intelligence(AI)commentators to explain the game live simultaneously.The real-time win rate prediction and key data characteristics of the game are important contents of the simultaneous interpretation.The exsiting works on winning rate prediction of HK foucs on historical game data driven techniques.Hence,the prediction rate and interpretability are hardly satisfactory.This paper uses the real combat data of HK,we propose a new Role and Real-time Syndication Network(RRSN)by learning from real HK combat data and heroes′interactions.Subsequently,through a large number of real data experiments to predict the real-time win rate of HK,the accuracy can archive 87%,and the proposed model is effective both in prediction accuracy and interpretability.Using the predictive model and algorithm proposed in this paper,AI commentators can give more accurate win rate predictions and analysis in the commentary,which greatly provides the experience of game audiences.