KARMA: Knowledge-Action Regularized Multimodal Alignment for Personalized Search at Taobao
📰 ArXiv cs.AI
KARMA aligns multimodal knowledge and actions for personalized search at Taobao
Action Steps
- Identify the knowledge-action gap in LLMs for personalized tasks
- Develop a multimodal alignment approach to bridge the gap
- Fine-tune LLMs with knowledge-action regularization for improved performance
- Deploy the KARMA model in a real-world personalized search system
Who Needs to Know This
AI engineers and researchers on a team can benefit from KARMA as it improves personalized search results, and product managers can utilize it to enhance user experience
Key Insight
💡 KARMA addresses the suboptimal performance of fine-tuned LLMs in industrial personalized tasks by aligning multimodal knowledge and actions
Share This
💡 KARMA bridges the knowledge-action gap in LLMs for personalized search
Key Takeaways
KARMA aligns multimodal knowledge and actions for personalized search at Taobao
Full Article
Title: KARMA: Knowledge-Action Regularized Multimodal Alignment for Personalized Search at Taobao
Abstract:
arXiv:2603.22779v1 Announce Type: cross Abstract: 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
Abstract:
arXiv:2603.22779v1 Announce Type: cross Abstract: 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
DeepCamp AI