Research map / Large language models / Reasoning and chain of thought
Cross-Scale Collaboration between LLMs and Lightweight Sequential Recommenders with Domain-Specific Latent Reasoning
Yipeng Zhang, Xin Wang, Hong Chen, Junwei Pan, Qian Li, Jun Zhang, Jie Jiang, Hong Mei, Wenwu Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a cross-scale collaboration framework between LLMs and lightweight sequential recommenders using domain-specific latent reasoning, addressing scarce recommendation data and LLM-based method limitations.
Problem
LLM-based sequential recommendation suffers from scarce recommendation-domain reasoning data and other limitations (the abstract is truncated after the first limitation), so LLMs are used to predict next items or generate semantic IDs.
Approach
Collaborates LLMs with lightweight sequential recommenders across scales, using domain-specific latent reasoning; exact mechanism not stated in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-scale framework (CoderRec) over EAGER-LLM | sequential recommender · Amazon Instruments | Recall@5: 0.0991 -> 0.1174 | Improved |
| add cross-scale framework (CoderRec) over EAGER-LLM | sequential recommender · Amazon Beauty | Recall@10: 0.0830 -> 0.0830 | No clear effect |
| keep item ID alongside semantic IDs (ablation) | CoderRec · Amazon Beauty | Recall@10: 0.0746 -> 0.0830 | Improved |
| add cross-scale reconstruction auxiliary loss (ablation) | CoderRec · Amazon Beauty | NDCG@10: 0.0495 -> 0.0511 | Improved |
| add domain-specific latent reasoning (ablation) | CoderRec · Amazon Beauty | Recall@10: 0.0806 -> 0.0830 | Improved |
| semantic IDs only (no item ID) vs TIGER (ablation) | CoderRec · Amazon Beauty | Recall@10: 0.0648 -> 0.0746 | Improved |
| increase latent-reasoning loss weight to 0.1 (ablation) | CoderRec · Amazon Beauty | Recall@10: 0.0830 -> 0.0787 | Hurt |
| vary auxiliary reconstruction weight lambda1 from 0.0 to 0.1 (ablation) | CoderRec · Amazon Beauty | Recall@10: 0.0812 -> 0.0830 | Improved |
| add LLM-derived semantic IDs (TIGER) over SASRec | sequential recommender · Amazon Beauty | Recall@5: 0.0387 -> 0.0454 | Improved |
| add latent reasoning to small model (ReaRec) over SASRec | SASRec · Amazon Beauty | Recall@5: 0.0387 -> 0.0541 | Improved |
Closest papers by meaning
- Reinforced Latent Reasoning for LLM-based Recommendation (ICLR 2026 · similarity 0.90)
- LLMRG: Improving Recommendations through Large Language Model Reasoning Graphs (AAAI 2024 · similarity 0.90)
- LLM-ESR: Large Language Models Enhancement for Long-tailed Sequential Recommendation (NeurIPS 2024 · similarity 0.89)
- LLM4RSR: Large Language Models as Data Correctors for Robust Sequential Recommendation (AAAI 2025 · similarity 0.89)
- SLMRec: Distilling Large Language Models into Small for Sequential Recommendation (ICLR 2025 · similarity 0.89)
- Semantic Convergence: Harmonizing Recommender Systems via Two-Stage Alignment and Behavioral Semantic Tokenization (AAAI 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
