平局处理是评测协议的一部分：重平局推荐分数的行序不变性审计
原标题：Tie Handling Is Part of the Evaluation Protocol: An Order-Invariance Audit for Tie-Heavy Recommender Scores
AI 摘要
这篇 arXiv 论文指出，推荐系统离线评测中当多个候选得分完全相同时，排序规则（tie-breaking）会成为评测协议的一部分。作者提出「行序不变性」概念，并在 Amazon Beauty & Personal Care 的 3 万行数据上验证：同一评分加权属性重叠分数在输入顺序打破平局时 NDCG@10 为 0.85，改用基于用户和物品 ID 的确定性哈希打破平局后降至 0.17；MovieLens Tag Genome 也呈现相同模式，而物品流行度分数几乎不受影响。论文推导了均匀随机打破平局下 Hit Rate 和 NDCG 的期望值，并给出六项报告清单。
正文节选
*1 Tie Handling Is Part of the Evaluation Protocol: An Order-Invariance Audit for Tie-Heavy Recommender Scores Abstract Offline top- evaluation often ranks one held-out relevant item together with sampled negatives. When several candidates receive exactly the same score, the tie-breaking rule becomes part of the ranking. A common implementation stores the relevant item first and then applies a stable sort, which preserves input order among equal scores; the relevant item therefore wins every tie