Reproduce “AgentLTV: An Agent-Based Unified Search-and-Evolution Framework for Automated Lifetime Value Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chaowei Wu, Huazhu Chen, Congde Yuan, Qirui Yang, Guoqing Song, Yue Gao, Li Luo, Frank Youhua Chen and 1 more
AgentLTV, an LLM-driven multi-agent system, automates the entire Lifetime Value (LTV) prediction pipeline by searching and evolving executable code to handle diverse data patterns and negative LTV values. It achieved an average 41.26% reduction in Error Rate and 15.44% improvement in Normalized Gini compared to strong baselines on an industrial dataset, demonstrating robust performance across various LTV segments in online deployment.
Key insights
- The combination of MCTS for broad exploration (identifying high-impact structural decisions) and EA for stable refinement is crucial for effectively navigating the LTV modeling search space.
- Optimal LTV models often emerge from a few high-impact structural decisions identified early in the search, following a "wide-but-shallow" exploration pattern.
- Explicitly handling negative LTV values through specialized output designs and calibration mechanisms significantly improves predictive accuracy and ranking quality, especially for high-risk user segments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.