# Aligning Multimodal Sequential Recommendations via Robust Direct Preference Optimization with Sparse MoE

arXiv: https://arxiv.org/abs/2603.29259 · HTML: https://mlsanity.com/paper/2603.29259/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Preference alignment for recommenders and RAG, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 17 of its 26 references that are in the map.
- First posted on arXiv 31 Mar 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Preference alignment for recommenders and RAG](https://mlsanity.com/alignment/preference-alignment-for-recommenders-and-rag/)
- Mixture-of-experts / [MoE for recommender systems](https://mlsanity.com/moe/moe-for-recommender-systems/)

## Builds on (most cited first, 17 of 17)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Self-Attentive Sequential Recommendation](https://mlsanity.com/paper/1808.09781/) (2018)
- [Neural Collaborative Filtering](https://mlsanity.com/paper/1708.05031/) (2017)
- [BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer](https://mlsanity.com/paper/1904.06690/) (2019)
- [Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering](https://mlsanity.com/paper/1602.01585/) (2016)
- [S^3-Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization](https://mlsanity.com/paper/2008.07873/) (2020)
- [RecBole: Towards a Unified, Comprehensive and Efficient Framework for Recommendation Algorithms](https://mlsanity.com/paper/2011.01731/) (2020)
- [Frequency Enhanced Hybrid Attention Network for Sequential Recommendation](https://mlsanity.com/paper/2304.09184/) (2023)
- [Non-invasive Self-attention for Side Information Fusion in Sequential Recommendation](https://mlsanity.com/paper/2103.03578/) (2021)
- [Linear Recurrent Units for Sequential Recommendation](https://mlsanity.com/paper/2310.02367/) (2023)
- [MISSRec: Pre-training and Transferring Multi-modal Interest-aware Sequence Representation for Recommendation](https://mlsanity.com/paper/2308.11175/) (2023)
- [IISAN: Efficiently Adapting Multimodal Representation for Sequential Recommendation with Decoupled PEFT](https://mlsanity.com/paper/2404.02059/) (2024)
- [Hierarchical Time-Aware Mixture of Experts for Multi-Modal Sequential Recommendation](https://mlsanity.com/paper/2501.14269/) (2025)
- [Sequence-level Semantic Representation Fusion for Recommender Systems](https://mlsanity.com/paper/2402.18166/) (2024)
- [Direct Preference Optimization for LLM-Enhanced Recommendation Systems](https://mlsanity.com/paper/2410.05939/) (2024)

## Abstract (arXiv)

Preference-based alignment objectives have been widely adopted, from RLHF-style pairwise learning in large language models to emerging applications in recommender systems. Yet, existing work rarely examines how Direct Preference Optimization (DPO) behaves under implicit feedback, where unobserved items are not reliable negatives. We conduct systematic experiments on multimodal sequential recommendation to compare common negative-selection strategies and their interaction with DPO training. Our central finding is that a simple modification, replacing deterministic hard negatives with stochastic sampling from a dynamic top-K candidate pool, consistently improves ranking performance. We attribute its effectiveness to two factors: (1) reducing erroneous suppressive gradients caused by false negatives, and (2) retaining informative hard signals while smoothing optimization via controlled stochasticity. With an optional sparse Mixture-of-Experts encoder for efficient capacity scaling, RoDPO achieves up to 5.25% NDCG@5 on three Amazon benchmarks, with nearly unchanged inference cost.