EVENT DOSSIER
To address the three major bottlenecks in MOBA games: difficult cold start, inconsistent data distribution, and lack of sufficient samples, researchers proposed the CHAMP (Cross-domain Hybrid Architecture for Matchmaking and Prediction) framework. This approach uses a cross-domain hybrid architecture, replacing traditional player profiles with a mixed set that integrates time-series and multi-mode statistical features, and constructs a domain-aware win rate network (DAWN). Offline tests showed that DAWN’s win rate prediction accuracy reached 67.73%, surpassing all baseline models; online A/B tests further confirmed that CHAMP reduced the “kill advantage” of players at lower levels by 20.73%, effectively alleviating the problem of matching imbalance.
CHAMP 框架解决了 MOBA 游戏匹配中冷启动、分布不一致及数据匮乏三大瓶颈。该研究提出跨域混合架构，将目标模式玩家画像替换为含时间序短序列与多模式统计的混合特征集，并构建域感知胜率网络（DAWN），通过域感知知识提取器与多编码器联合学习模式条件表征。离线测试中 DAWN 胜率预测准确率达 67.73%，优于所有基线模型；在线 A/B 测试显示，CHAMP 使低段位玩家 5 分钟击杀碾压率降低最高 20.73%，有效减少了不平衡匹配。