LLMDistill4Ads: Using Cross-Encoders to Distill from LLM Signals for Advertiser Keyphrase Recommendations
This work addresses the challenge of providing relevant keyphrase suggestions for e-commerce advertisers to enhance advertising campaigns and maintain system integrity, though it appears incremental as it builds on existing distillation and LLM methods.
The paper tackled the problem of mitigating click-induced biases in advertiser keyphrase recommendations on an e-commerce platform by using a distillation framework with an LLM teacher, cross-encoder assistant, and bi-encoder student model, resulting in improved alignment with seller, search, and buyer judgments.
E-commerce sellers are advised to bid on keyphrases to boost their advertising campaigns. These keyphrases must be relevant to prevent irrelevant items from cluttering search systems and to maintain positive seller perception. It is vital that keyphrase suggestions align with seller, search and buyer judgments. Given the challenges in collecting negative feedback in these systems, LLMs have been used as a scalable proxy to human judgments. This paper presents an empirical study on a major ecommerce platform of a distillation framework involving an LLM teacher, a cross-encoder assistant and a bi-encoder Embedding Based Retrieval (EBR) student model, aimed at mitigating click-induced biases in keyphrase recommendations.