SP
BravenNow
From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation in Production
| USA | technology | ✓ Verified - arxiv.org

From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation in Production

#Large Language Models #Verbalization #Recommendation Systems #Reinforcement Learning #User Interaction Logs #Natural Language Processing #Data-Centric Framework

📌 Key Takeaways

  • Researchers developed a novel framework for optimizing verbalization in LLM-based recommendation systems
  • The approach uses reinforcement learning to transform raw user interaction logs into optimized textual contexts
  • System achieved up to 93% improvement in recommendation accuracy over traditional methods
  • Framework learns to filter noise, incorporate metadata, and reorganize information effectively

📖 Full Retelling

A team of researchers led by Yucheng Shi and including 10 other authors introduced a novel framework on February 24, 2026, that addresses the challenge of converting structured user interaction logs into effective natural language inputs for large language model-based recommendation systems, aiming to overcome the limitations of existing rigid template methods that produce suboptimal recommendation results. The paper, titled 'From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation in Production,' presents a data-centric approach that utilizes reinforcement learning to create a verbalization agent capable of transforming raw user interaction histories into optimized textual contexts. This innovative method uses recommendation accuracy as the training signal, enabling the system to learn how to filter out noise, incorporate relevant metadata, and reorganize information to enhance downstream predictions. The researchers tested their framework on a large-scale industrial streaming dataset and found that the learned verbalization approach delivered up to 93% relative improvement in discovery item recommendation accuracy compared to traditional template-based baselines. Further analysis revealed emergent strategies such as user interest summarization, noise removal, and syntax normalization, providing valuable insights into effective context construction for LLM-based recommender systems in production environments.

🏷️ Themes

Artificial Intelligence, Recommendation Systems, Natural Language Processing

📚 Related People & Topics

Reinforcement learning

Reinforcement learning

Field of machine learning

In machine learning and optimal control, reinforcement learning (RL) is concerned with how an intelligent agent should take actions in a dynamic environment in order to maximize a reward signal. Reinforcement learning is one of the three basic machine learning paradigms, alongside supervised learnin...

View Profile → Wikipedia ↗

Large language model

Type of machine learning model

A large language model (LLM) is a language model trained with self-supervised machine learning on a vast amount of text, designed for natural language processing tasks, especially language generation. The largest and most capable LLMs are generative pre-trained transformers (GPTs) that provide the c...

View Profile → Wikipedia ↗

Verbalisation

Verbalisation (or verbalization, see spelling differences) is a process by which different psychological events in an individual are made in verbal form, i.e. described "in their own words". According to psychoanalytic concepts, only when clients manage to verbalise their own experiences or problems...

View Profile → Wikipedia ↗

Entity Intersection Graph

Connections for Reinforcement learning:

🌐 Large language model 7 shared
🌐 Artificial intelligence 6 shared
🌐 Machine learning 4 shared
🏢 Science Publishing Group 2 shared
🌐 Reasoning model 2 shared
View full profile
Original Source
--> Computer Science > Artificial Intelligence arXiv:2602.20558 [Submitted on 24 Feb 2026] Title: From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation in Production Authors: Yucheng Shi , Ying Li , Yu Wang , Yesu Feng , Arjun Rao , Rein Houthooft , Shradha Sehgal , Jin Wang , Hao Zhen , Ninghao Liu , Linas Baltrunas View a PDF of the paper titled From Logs to Language: Learning Optimal Verbalization for LLM-Based Recommendation in Production, by Yucheng Shi and 10 other authors View PDF HTML Abstract: Large language models 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. Further analysis reveals emergent strategies such as user interest summarization, noise removal, and syntax normalization, offering insights into effective context construction for LLM-based recommender systems. Comments: Work in progress Subjects: Artificial Intelligence (cs.AI) ; Information Retrieval (cs.IR) Cite as: arXiv:2602.20558 [cs.AI] (or arXiv:2602.20558v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2602.20558 Focus to learn more arXiv-issued DOI via DataCite (pending registrat...
Read full article at source

Source

arxiv.org

More from USA

News from Other Countries

🇬🇧 United Kingdom

🇺🇦 Ukraine