Multiplayer Online Battle Arena (MOBA) games rely on matchmaking to keep competitive balance. Our previous system CUPID treated matchmaking as an assignment re‑optimization problem and showed that a single‑mode win‑rate predictor can noticeably rebalance teams. Deploying such a system across diverse player populations, however, reveals three practical bottlenecks: most queuing players lack sufficient in‑mode match history (cold start), skill distributions shift dramatically across rank tiers (distribution inconsistency), and extreme skill segments suffer severe data starvation.
CHAMP introduces a cross‑domain matchmaking framework to address these bottlenecks. To mitigate data sparsity and cold starts, CHAMP replaces the target‑mode‑only player profile with a hybrid domain feature set: a timestamp‑ordered cross‑mode short‑term sequence whose slices are annotated with target‑domain features, together with per‑mode breakdowns of long‑term, real‑time, and team statistics.
We further propose the Domain‑Aware Win‑rate Network (DAWN). A Domain‑Aware Knowledge Extractor (DAKE) compiles target‑mode attributes into learnable representations that feed Domain‑Aware Temporal/Spatial/Permutation OmniNet Encoders (DATOE, DASOE, DAPOE), enabling mode‑conditioned representations and per‑mode debiasing to be learned jointly within a single shared network.
Online, one trained DAWN serves every supported mode, with per‑mode position‑satisfaction thresholds as the only mode‑specific knob. Offline evaluation shows DAWN achieves 67.73% win‑rate prediction accuracy, outperforming all evaluated attention and sequence baselines.
In large‑scale A/B tests covering the entire ladder of a major MOBA title, CHAMP consistently reduces imbalanced matches. For lower‑tier players, the 5‑minute kill‑crushing rate drops by up to 20.73%.
Review