LACRM: LLM-Guided Mining of Directed Advertiser Competition Relationships in E-Commerce Advertising

Researchers propose LACRM, a framework that uses large language models to build directed competition graphs with continuous edge intensities for e-commerce advertising. The three-stage pipeline employs LLM-driven role-separated profiling to enable asymmetric multi-route retrieval of both realized and latent rivals, and incorporates listwise learning-to-rank aligned with temporal auction behavior to model directed competition intensity.