# KARMA: Knowledge-Action Regularized Multimodal Alignment for Personalized Search at Taobao

arXiv: https://arxiv.org/abs/2603.22779 · HTML: https://mlsanity.com/paper/2603.22779/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 11 of its 16 references that are in the map.
- First posted on arXiv 24 Mar 2026.

## Builds on (most cited first, 11 of 11)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Self-Attentive Sequential Recommendation](https://mlsanity.com/paper/1808.09781/) (2018)
- [Wide & Deep Learning for Recommender Systems](https://mlsanity.com/paper/1606.07792/) (2016)
- [Deep Interest Network for Click-Through Rate Prediction](https://mlsanity.com/paper/1706.06978/) (2017)
- [HLLM: Enhancing Sequential Recommendations via Hierarchical Large Language Models for Item and User Modeling](https://mlsanity.com/paper/2409.12740/) (2024)
- [Sparse Meets Dense: Unified Generative Recommendations with Cascaded Sparse-Dense Representations](https://mlsanity.com/paper/2503.02453/) (2025)
- [NoteLLM: A Retrievable Large Language Model for Note Recommendation](https://mlsanity.com/paper/2403.01744/) (2024)
- [Enhancing Taobao Display Advertising with Multimodal Representations: Challenges, Approaches and Insights](https://mlsanity.com/paper/2407.19467/) (2024)
- [LEARN: Knowledge Adaptation from Large Language Model to Recommendation for Practical Industrial Application](https://mlsanity.com/paper/2405.03988/) (2024)
- [STAR: A Simple Training-free Approach for Recommendations using Large Language Models](https://mlsanity.com/paper/2410.16458/) (2024)
- [HHFT: Hierarchical Heterogeneous Feature Transformer for Recommendation Systems](https://mlsanity.com/paper/2511.20235/) (2025)

## Built on by (most cited first, 2 of 2)

- [Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation](https://mlsanity.com/paper/2604.10098/) (2026)
- [Personalization Meets Safety:Mechanisms,Risks,and Mitigations in Personalized LLMs](https://mlsanity.com/paper/2606.09038/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) are equipped with profound semantic knowledge, making them a natural choice for injecting semantic generalization into personalized search systems. However, in practice we find that directly fine-tuning LLMs on industrial personalized tasks (e.g. next item prediction) often yields suboptimal results. We attribute this bottleneck to a critical Knowledge--Action Gap: the inherent conflict between preserving pre-trained semantic knowledge and aligning with specific personalized actions by discriminative objectives. Empirically, action-only training objectives induce Semantic Collapse, such as attention "sinks". This degradation severely cripples the LLM's generalization, failing to bring improvements to personalized search systems. We propose KARMA (Knowledge--Action Regularized Multimodal Alignment), a unified framework that treats semantic reconstruction as a train-only regularizer. KARMA optimizes a next-interest embedding for retrieval (Action) while enforcing semantic decodability (Knowledge) through two complementary objectives: (i) history-conditioned semantic generation, which anchors optimization to the LLM's native next-token distribution, and (ii) embedding-conditioned semantic reconstruction, which constrains the interest embedding to remain semantically recoverable. On Taobao search system, KARMA mitigates semantic collapse (attention-sink analysis) and improves both action metrics and semantic fidelity. In ablations, semantic decodability yields up to +22.5 HR@200. With KARMA, we achieve +0.25 CTR AUC in ranking, +1.86 HR in pre-ranking and +2.51 HR in recalling. Deployed online with low inference overhead at ranking & pre-ranking stage, KARMA drives +0.9% increase in GMV.