BROWSE
Detail
Conference Paper
다중 팀 게임 랭킹 예측을 위한 개인 단위 집계와 팀 단위 학습의 비교
Cited - time in 
Share 
Share
- Issue Date
- 2026-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2026, pp.1-4
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- Predicting final placements from intermediate states in multi-team games involves considering the relative order of teams at a given game state. This study formulates final placement prediction in Eternal Return as a learning-to-rank problem and examines performance differences associated with input representation. We first compare a ranking model with regression-based baselines and adopt the ranking model for subsequent analysis. Using this model, we perform game-level 10-fold cross-validation to evaluate two input representations under a unified experimental setting, keeping the source data and evaluation protocol consistent across representations. The results show that preserving player-level states as separate instances achieves higher NDCG and Hit than aggregated team-level representations. These findings suggest that learning to rank is well suited to this task and that input representation can lead to performance differences even with identical source data.
- KSP Keywords
- Cross validation(CV), Evaluation Protocol, Intermediate states, Performance difference, Placement prediction, Ranking model, Regression-based, Relative order, game state, input representation, learning to rank