# Scaling Articulated Rationales for MLLM-based Recommendation

arXiv: https://arxiv.org/abs/2609.17639 · HTML: https://mlsanity.com/paper/2609.17639/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 32 references that are in the map.
- First posted on arXiv 15 Sep 2026.

## Builds on (most cited first, 24 of 24)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation](https://mlsanity.com/paper/2402.03216/) (2024)
- [STaR: Bootstrapping Reasoning With Reasoning](https://mlsanity.com/paper/2203.14465/) (2022)
- [Prometheus: Inducing Fine-grained Evaluation Capability in Language Models](https://mlsanity.com/paper/2310.08491/) (2023)
- [TALLRec: An Effective and Efficient Tuning Framework to Align Large Language Model with Recommendation](https://mlsanity.com/paper/2305.00447/) (2023)
- [JudgeBench: A Benchmark for Evaluating LLM-based Judges](https://mlsanity.com/paper/2410.12784/) (2024)
- [Agent-as-a-Judge: Evaluate Agents with Agents](https://mlsanity.com/paper/2410.10934/) (2024)
- [V-STaR: Training Verifiers for Self-Taught Reasoners](https://mlsanity.com/paper/2402.06457/) (2024)
- [Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge](https://mlsanity.com/paper/2506.21506/) (2025)
- [LLaRA: Large Language-Recommendation Assistant](https://mlsanity.com/paper/2312.02445/) (2023)
- [MiniCPM4: Ultra-Efficient LLMs on End Devices](https://mlsanity.com/paper/2506.07900/) (2025)
- [Agentic Reward Modeling: Integrating Human Preferences with Verifiable Correctness Signals for Reliable Reward Systems](https://mlsanity.com/paper/2502.19328/) (2025)
- [LiveStar: Live Streaming Assistant for Real-World Online Video Understanding](https://mlsanity.com/paper/2511.05299/) (2025)
- [Review-based Recommender Systems: A Survey of Approaches, Challenges and Future Perspectives](https://mlsanity.com/paper/2405.05562/) (2024)
- [RecGPT-V2 Technical Report](https://mlsanity.com/paper/2512.14503/) (2025)
- [LiViBench: An Omnimodal Benchmark for Interactive Livestream Video Understanding](https://mlsanity.com/paper/2601.15016/) (2026)
- [KuaiLive: A Real-time Interactive Dataset for Live Streaming Recommendation](https://mlsanity.com/paper/2508.05633/) (2025)
- [OneLive: Dynamically Unified Generative Framework for Live-Streaming Recommendation](https://mlsanity.com/paper/2602.08612/) (2026)
- [SARM: LLM-Augmented Semantic Anchor for End-to-End Live-Streaming Ranking](https://mlsanity.com/paper/2602.09401/) (2026)
- [LiveLongBench: Tackling Long-Context Understanding for Spoken Texts from Live Streams](https://mlsanity.com/paper/2504.17366/) (2025)
- [KuaiLive-M3: A Multi-Modal, Multi-Domain, and Multi-Feedback Dataset for Live Streaming Recommendation](https://mlsanity.com/paper/2607.24862/) (2026)

## Abstract (arXiv)

We presented SARA, an industrial framework that transforms sparse articulated user rationales into scalable recommendation signals. Its data engine curates questionnaire responses into SARA-HQ, providing explicit preference supervision for aligning SARA-7B through SFT and Quality-Refining DPO. This alignment extends rationale generation from 86,564 questionnaire-covered authors to the full 10M-author space. SARA-Ranker translates the generated positive and negative rationales into features for user--author interaction modeling and negative-feedback history modeling, connecting articulated reasons to production ranking. Evaluation on unseen authors demonstrates that SARA-7B generates more specific, relevant, and grounded rationales than the evaluated general-purpose MLLMs. On top of a strong industrial ranking baseline with multimodal features, separate online A/B tests show that positive-rationale integration increases watch time by 0.99%, while negative-rationale integration reduces Hate feedback by 8.16%. Daily refresh and more than 30 days of production deployment further demonstrate the operational feasibility of the approach. These findings establish articulated rationales as a useful complement to behavioral and content signals, and demonstrate a practical role for MLLMs in scaling sparse human explanations into preference information that improves industrial recommendation.