• Computer Science > Artificial Intelligence [Submitted on 24 Feb 2026] Title:From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation in Production View PDF HTML (experimental)Abstract:Large language models (LLMs) are promising backbones for generative recommender systems, yet a key challenge remains underexplored: verbalization, i.e., converting structured user interaction logs into effective natural language inputs. • Existing methods rely on rigid templates that simply concatenate fields, yielding suboptimal representations for recommendation. • We propose a data-centric framework that learns verbalization for LLM-based recommendation. • Using reinforcement learning, a verbalization agent transforms raw interaction histories into optimized textual contexts, with recommendation accuracy as the training signal. • This agent learns to filter noise, incorporate relevant metadata, and reorganize information to improve downstream predictions. • Experiments on a large-scale industrial streaming dataset show that learned verbalization delivers up to 93% relative improvement in discovery item recommendation accuracy over template-based baselines.

Article Summaries:

  • Researchers have introduced a data‑centric framework that learns to convert raw user interaction logs into optimized natural‑language contexts for large‑language‑model (LLM) recommender systems. Using reinforcement learning, a verbalization agent refines interaction histories-filtering noise, selecting relevant metadata, and reorganizing information-so that the LLM receives higher‑quality input. Applied to a large industrial streaming dataset, the learned verbalization achieved up to a 93 % relative improvement in discovery‑item recommendation accuracy compared with conventional template‑based approaches. Analysis revealed emergent strategies such as user‑interest summarization, noise removal, and syntax normalization, offering new insights into effective context construction for LLM‑driven recommendations.

Sources: