Accepted to MoFA @ ICML 2025
Authors: Chloe Ho, Ishneet Sukhvinder Singh, Diya Sharma, Tanvi Reddy Anumandla
Search algorithms and user query relevance have given LLMs the ability to return relevant information, but the effect of content phrasing on ad visibility remains underexplored. We explore how rewriting advertisements through language models can enhance their ranking in retrieval systems and boost their presence in LLM-generated responses, without altering the retrieval model itself. We introduce a supervised fine-tuning framework with a custom loss balancing semantic relevance and content fidelity. To evaluate effectiveness, we propose two metrics: DeltaMRR@K (ranking improvement) and DeltaDIR@K (inclusion frequency improvement). Experiments across both instruction-based and few-shot prompting demonstrate that PPO-trained models substantially outperform prompt engineering and supervised fine-tuning approaches, underscoring the significance of ad phrasing and reinforcement learning in retrieval-integrated LLM systems.
