LLM-EDT: Dual-Phase Training for Cross-Domain Recommendation
A new method called LLM-EDT (Large Language Model Enhanced Cross-domain Sequential Recommendation with Dual-phase Training) addresses two key challenges in Cross-domain Sequential Recommendation (CDSR): the imbalance issue, where one domain's interactions dominate, and the transition issue, which hinders capturing cross-domain preferences. Current LLM-enhanced CDSR methods fail to recognize irrelevant noise and suffer from rough profiling. LLM-EDT uses large language models as generators and encoders to partially alleviate these problems. The paper is published on arXiv with ID 2511.19931.
Key facts
- LLM-EDT stands for Large Language Model Enhanced Cross-domain Sequential Recommendation with Dual-phase Training.
- The method addresses imbalance and transition issues in CDSR.
- Imbalance issue: one domain's interactions dominate user behavior.
- Transition issue: difficulty capturing cross-domain preferences in mixed sequences.
- LLMs are used as generators and encoders to alleviate these issues.
- Current LLM-enhanced CDSR methods fail to recognize irrelevant noise.
- Current methods also suffer from rough profiling problems.
- The paper is available on arXiv with ID 2511.19931.
Entities
Institutions
- arXiv