One-step Reach: LLM-based Keyword Generation for Sponsored Search Advertising

Yang Wang, Zheyi Sha, Kunhai Lin, Chaobing Feng, Kunhong Zhu, Lipeng Wang, Xuewu Jiao, Fei Huang, Chao Ye, Dengwu He, Zhi Guo, Shuanglong Li, Lin Liu

Abstract: Query keyword matching plays a crucial role in sponsored search advertising by retrieving semantically related keywords of the user query to target relevant advertisements. Conventional technical solutions adopt the retrieve-judge-then-rank retrieval framework structured in cascade funnels. However, it has limitations in accurately depicting the semantic relevance between the query and keyword, and the cumulative funnel losses result in unsatisfactory precision and recall. To address the above issues, this paper proposes a Large Language Model (LLM)-based keyword generation method (LKG) to reach related keywords from the search query in one step. LKG models the query keyword matching as an end-to-end keyword generation task based on the LLM through multi-match prompt tuning. Moreover, it employs the feedback tuning and the prefix tree-based constrained beam search to improve the generation quality and efficiency. Extensive offline experiments and online A/B testing demonstrate the effectiveness and superiority of LKG which is fully deployed in the Baidu sponsored search system bringing significant improvements.