Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

KP-Agent: Keyword Pruning in Sponsored Search Advertising via LLM-Powered Contextual Bandits

Hou-Wan Long, Yicheng Song, Zidong Wang, Tianshu Sun

TL;DR

This paper tackles keyword pruning in sponsored search advertising by framing it as a contextual bandit problem and introducing KP-Agent, an LLM-based agent augmented with SSA-domain tools and a memory module. The system uses domain-specific tools to process tabular SSA data without hallucination, a memory-augmented reflection mechanism for few-shot learning, and code generation to enact pruning decisions, all guided by advertiser-side KPIs. Empirical evaluation on 21 days of Meituan SSA data shows KP-Agent achieves up to 49.28% higher cumulative profit than baseline pruning strategies, with larger gains when the minimum keyword retention is lower. The work demonstrates a practical, adaptive pruning approach that improves resource allocation under budget constraints and demonstrates significant potential for real-world SSA optimization.

Abstract

Sponsored search advertising (SSA) requires advertisers to constantly adjust keyword strategies. While bid adjustment and keyword generation are well-studied, keyword pruning-refining keyword sets to enhance campaign performance-remains under-explored. This paper addresses critical inefficiencies in current practices as evidenced by a dataset containing 0.5 million SSA records from a pharmaceutical advertiser on search engine Meituan, China's largest delivery platform. We propose KP-Agent, an LLM agentic system with domain tool set and a memory module. By modeling keyword pruning within a contextual bandit framework, KP-Agent generates code snippets to refine keyword sets through reinforcement learning. Experiments show KP-Agent improves cumulative profit by up to 49.28% over baselines.

KP-Agent: Keyword Pruning in Sponsored Search Advertising via LLM-Powered Contextual Bandits

TL;DR

This paper tackles keyword pruning in sponsored search advertising by framing it as a contextual bandit problem and introducing KP-Agent, an LLM-based agent augmented with SSA-domain tools and a memory module. The system uses domain-specific tools to process tabular SSA data without hallucination, a memory-augmented reflection mechanism for few-shot learning, and code generation to enact pruning decisions, all guided by advertiser-side KPIs. Empirical evaluation on 21 days of Meituan SSA data shows KP-Agent achieves up to 49.28% higher cumulative profit than baseline pruning strategies, with larger gains when the minimum keyword retention is lower. The work demonstrates a practical, adaptive pruning approach that improves resource allocation under budget constraints and demonstrates significant potential for real-world SSA optimization.

Abstract

Sponsored search advertising (SSA) requires advertisers to constantly adjust keyword strategies. While bid adjustment and keyword generation are well-studied, keyword pruning-refining keyword sets to enhance campaign performance-remains under-explored. This paper addresses critical inefficiencies in current practices as evidenced by a dataset containing 0.5 million SSA records from a pharmaceutical advertiser on search engine Meituan, China's largest delivery platform. We propose KP-Agent, an LLM agentic system with domain tool set and a memory module. By modeling keyword pruning within a contextual bandit framework, KP-Agent generates code snippets to refine keyword sets through reinforcement learning. Experiments show KP-Agent improves cumulative profit by up to 49.28% over baselines.
Paper Structure (11 sections, 3 equations, 4 figures, 1 table, 1 algorithm)

This paper contains 11 sections, 3 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Motivation
  3. Problem Formulation
  4. KP-Agent's Solution Paradigm
  5. SSA Domain-Specialized Tools for Hallucination-Free Tabular Processing
  6. Memory-Augmented Reflection for Few-Shot Learning
  7. Experiment
  8. Dataset
  9. Experiment Setup
  10. Main Result
  11. Conclusion

Figures (4)

  • Figure 1: Keyword Pruning and KP-Agent.
  • Figure 2: Empirical evidence from Meituan highlighting the need of effective keyword pruning method
  • Figure 3: Simplified Demonstration of Tool and Code Snippet
  • Figure 4: Performance comparison between KP-Agent and baseline methods exhibiting daily profit (first row) and cumulative profit (second row) across varying $N_{\text{min}}$ values, where the different color represent the corresponding methods: $\blacksquare$ KP-Agent, $\blacksquare$ Impression-Rank, $\blacksquare$ CTR-Rank, $\blacksquare$ CVR-Rank, $\blacksquare$ Impression Regression.