YMM-TECH Algorithm Competition Final: Problem Background, Evaluation Methodology, and Scoring Details

The YMM-TECH algorithm competition final was held in mid‑September at Nanjing University of Posts and Telecommunications, organized by Manbang Group (Yunmanman). The event introduced the competition’s final problem and its underlying logic.

In today’s sharing‑economy context, the "Internet + logistics" model suffers from low efficiency and high empty‑run rates. The competition’s task is to mine drivers’ historical behavior to personalize cargo recommendations, enabling faster and more accurate driver‑cargo matching and improving road‑logistics efficiency.

The primary evaluation method is ranking accuracy. Based on a survey of recent ACM Transactions on Information Systems papers on recommendation systems, the organizers adopt a position‑weighted metric where utility decays with rank (α‑decay). This approach rewards higher‑ranked correct recommendations and penalizes lower‑ranked ones.

The scoring formula uses rank(i) to denote the position of item i in a driver’s recommendation list and rankscore_u^{max} as the maximum possible score for driver u. The overall driver score is calculated as the percentage of correctly predicted clicks over the total predicted clicks, normalized by the maximum rank score.

An illustrative example shows a driver’s true click list, the submitted prediction set with scores, the computation of maximal scores for each driver, and the final normalized ranking score. The example images detail each step of the calculation.

Each driver’s recommendation list is limited to 20 cargo items, based on analysis of click‑through distribution (average of 18 items). All data used in the competition have been anonymized.

The competition spanned four months; the winning team came from Nanjing University of Posts and Telecommunications, with runners‑up from Shanghai Jiao Tong University, Zhejiang University, Tongji University, USTC, Shanghai Maritime University, and UCLA. Manbang Group plans to continue leveraging technology to further improve road‑logistics efficiency.