Information Flow Advertising Relevance Ranking Method Using Knowledge-Enhanced BERT

Efficient semantic relevance ranking is essential for intelligent information delivery in modern wireless communication and data-driven service systems, where accurate content matching can improve the utilization of communication resources and user experience. To overcome the limitations of conventional BERT models in semantic understanding, business knowledge integration, and real-time inference, this study proposes a Knowledge-Enhanced BERT (KE-BERT) framework for information flow advertising relevance ranking. The proposed approach incorporates multi-source domain knowledge into the Transformer attention mechanism through entity alignment and further employs a dual-channel interaction architecture with adaptive gating to dynamically balance semantic and knowledge representations. To satisfy industrial latency requirements, layer pruning, quantization, and knowledge distillation are jointly adopted to construct a lightweight inference model. Experimental results demonstrate that the proposed framework achieves superior ranking performance and significantly improves online efficiency, yielding an NDCG@10 of 0.782 with a P99 latency of only 35 ms while increasing click-through rate and revenue-related metrics in real-world deployment. The proposed knowledge-aware semantic ranking strategy provides a practical reference for intelligent information dissemination and semantic resource optimization in next-generation wireless communication and electromagnetic information service systems.