CARTS: Collaborative Agents for Recommendation Textual Summarization

2025-06-21

AI Summary

Generating highly relevant, strictly length-constrained summary titles for item sets remains challenging in recommender systems: existing LLM-based summarization methods struggle to simultaneously ensure semantic coherence and hard token-length constraints. This paper proposes a multi-agent large language model framework tailored for recommendation scenarios, featuring a three-stage collaborative mechanism—feature-aware generation enhancement, feedback-driven iterative refinement, and multi-agent consensus arbitration—to jointly optimize title relevance and conciseness. The method integrates item feature extraction, interpretable refinement, and dynamic length control. Evaluated on a large-scale e-commerce dataset and online A/B tests, our approach achieves a 23.6% improvement in title relevance over single-pass generation and chain-of-thought baselines, with statistically significant gains in user click-through rate (p < 0.01), demonstrating its effectiveness and practicality in real-world recommendation deployments.

Technology Category

• Natural Language Processing: Summarization
• Multiagent Systems: Mechanism Design
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Current recommendation systems often require some form of textual data summarization, such as generating concise and coherent titles for product carousels or other grouped item displays. While large language models have shown promise in NLP domains for textual summarization, these approaches do not directly apply to recommendation systems, where explanations must be highly relevant to the core features of item sets, adhere to strict word limit constraints. In this paper, we propose CARTS (Collaborative Agents for Recommendation Textual Summarization), a multi-agent LLM framework designed for structured summarization in recommendation systems. CARTS decomposes the task into three stages-Generation Augmented Generation (GAG), refinement circle, and arbitration, where successive agent roles are responsible for extracting salient item features, iteratively refining candidate titles based on relevance and length feedback, and selecting the final title through a collaborative arbitration process. Experiments on large-scale e-commerce data and live A/B testing show that CARTS significantly outperforms single-pass and chain-of-thought LLM baselines, delivering higher title relevance and improved user engagement metrics.

Problem

• Summarizing textual data for recommendation systems effectively
• Ensuring summaries align with item features and length limits
• Improving title relevance and user engagement in recommendations

Innovation

• Multi-agent LLM framework for structured summarization
• Three-stage process: GAG, refinement, arbitration
• Enhances title relevance and user engagement