Meituan Technical Team's Three Papers Accepted at SIGIR 2024: Ad Auction Integration, Federated Recommendation, and POI Recommendation

This article presents three papers from Meituan's technical team accepted at SIGIR 2024, covering ad auction integration, federated recommendation, and POI recommendation.

The first paper, "Deep Automated Mechanism Design for Integrating Ad Auction and Allocation in Feed," proposes a deep learning‑based mechanism that jointly models ad auction and list‑wise mixing, accounts for positional externalities, and ensures incentive compatibility while boosting ad revenue and GMV in offline and online experiments.

The second paper, "ReFer: Retrieval‑Enhanced Vertical Federated Recommendation for Full Set User Benefit," extends federated learning to include non‑overlapping users, introduces the ReFer framework with a two‑stage distributed retrieval and attention‑fusion mechanism to mitigate missing cross‑domain features and interest bias, improving recommendation performance for all users.

The third paper, "Disentangled Contrastive Hypergraph Learning for Next POI Recommendation," introduces DCHL, which decouples user‑POI interactions into collaborative, transfer, and geographic views, learns view‑specific hypergraph representations, fuses them adaptively, and uses cross‑view contrastive learning to capture complementary signals, achieving superior POI recommendation on real‑world datasets.

The work includes collaborations with Tsinghua University, IIE CAS, UCAS, and others, and an online talk is scheduled for July 11, 2024.