In e-commerce advertising, platforms often optimize each advertiser's decisions in isolation, according to a new paper on arXiv (2610.13959). The authors argue that this misses a key factor: an advertiser's real-world performance is heavily influenced by rivals competing for the same users. Without modeling these externalities, ad allocation and auction strategies may be suboptimal.

The paper introduces LACRM, an LLM-guided approach to mining directed advertiser competition relationships. Rather than relying on static or hand-crafted signals, the method leverages large language models to infer the structure of rivalry from available data. The directed nature of the relationships suggests that competition is not necessarily symmetric—one advertiser may be more affected by another than the reverse.

While the abstract does not provide experimental details, the core idea is clear: making competition explicit could give ad platforms a more realistic view of the marketplace. If LACRM works as described, it could inform better bidding, budget allocation, and auction design in e-commerce advertising.