![[2602.16932] RankEvolve: Automating the Discovery of Retrieval Algorithms via LLM-Driven Evolution](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16932] RankEvolve: Automating the Discovery of Retrieval Algorithms via LLM-Driven Evolution
Summary
The paper presents RankEvolve, a novel approach utilizing large language models (LLMs) to automate the discovery of retrieval algorithms, enhancing performance over traditional methods.
Why It Matters
As information retrieval remains crucial in various applications, automating the discovery of effective algorithms can significantly improve efficiency and effectiveness in processing vast datasets. This research highlights the potential of LLMs in advancing algorithm development, which could lead to breakthroughs in search technology.
Key Takeaways
- RankEvolve uses LLMs to automate the discovery of retrieval algorithms.
- The approach improves upon traditional methods like BM25 through evolutionary search.
- Evolved algorithms demonstrate effectiveness across multiple information retrieval datasets.
- The study suggests a practical path for automatic algorithm discovery in AI.
- Evaluator-guided evolution can lead to novel and efficient ranking algorithms.
Computer Science > Information Retrieval arXiv:2602.16932 (cs) [Submitted on 18 Feb 2026] Title:RankEvolve: Automating the Discovery of Retrieval Algorithms via LLM-Driven Evolution Authors:Jinming Nian, Fangchen Li, Dae Hoon Park, Yi Fang View a PDF of the paper titled RankEvolve: Automating the Discovery of Retrieval Algorithms via LLM-Driven Evolution, by Jinming Nian and 3 other authors View PDF HTML (experimental) Abstract:Retrieval algorithms like BM25 and query likelihood with Dirichlet smoothing remain strong and efficient first-stage rankers, yet improvements have mostly relied on parameter tuning and human intuition. We investigate whether a large language model, guided by an evaluator and evolutionary search, can automatically discover improved lexical retrieval algorithms. We introduce RankEvolve, a program evolution setup based on AlphaEvolve, in which candidate ranking algorithms are represented as executable code and iteratively mutated, recombined, and selected based on retrieval performance across 12 IR datasets from BEIR and BRIGHT. RankEvolve starts from two seed programs: BM25 and query likelihood with Dirichlet smoothing. The evolved algorithms are novel, effective, and show promising transfer to the full BEIR and BRIGHT benchmarks as well as TREC DL 19 and 20. Our results suggest that evaluator-guided LLM program evolution is a practical path towards automatic discovery of novel ranking algorithms. Subjects: Information Retrieval (cs.IR); Artificial I...
