Research map / Large language models / Language, safety and interpretability
LLM-Aligned Geographic Item Tokenization for Local-Life Recommendation
Hao Jiang, Guoquan Wang, Donglin Zhou, Sheng Yu, Yang Zeng, Wencong Zeng, Kun Gai, Guorui Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add LGSID tokenization to TWIN | TWIN · Kuaishou local-life | AUC: 0.5898 -> 0.6263 | Improved |
| add LGSID tokenization to DIN vs RQ-VAE | DIN · Kuaishou local-life | AUC: 0.6185 -> 0.6276 | Improved |
| add LGSID tokenization to DIEN vs Res-KMeans | DIEN · Kuaishou local-life | AUC: 0.6369 -> 0.6484 | Improved |
| use LGSID instead of RQ-VAE in TIGER | TIGER · Kuaishou local-life | Hit@10: 0.3880 -> 0.5077 | Improved |
| use LGSID instead of RQ-VAE in OneRec | OneRec · Kuaishou local-life | Hit@10: 0.4534 -> 0.5537 | Improved |
| use Lin et al. quantization in TIGER | TIGER · Kuaishou local-life | Hit@10: 0.3880 -> 0.2067 | Hurt |
| use list-wise reward instead of point-wise reward (ablation) | G-DPO · Geographic alignment eval | Town coverage T@5: 0.1452 -> 0.5584 | Improved |
| add density-aware list-wise sampling (ablation) | G-DPO · Geographic alignment eval | Town coverage T@5: 0.5584 -> 0.6114 | Improved |
| add domain mixed preference pairs (ablation) | G-DPO · Geographic alignment eval | Semantic similarity Top@5: 0.7411 -> 0.8107 | Improved |
| add domain mixed preference pairs (ablation) | G-DPO · Geographic alignment eval | Town coverage T@5: 0.6114 -> 0.1816 | Hurt |
| add similarity regularization (ablation) | G-DPO · Geographic alignment eval | Semantic similarity Top@5: 0.8107 -> 0.8856 | Improved |
| apply G-DPO alignment of the LLM (ablation) | LGSID · Geographic alignment eval | Province coverage P@5: 0.8716 -> 0.9905 | Improved |
Closest papers by meaning
- Semantic Convergence: Harmonizing Recommender Systems via Two-Stage Alignment and Behavioral Semantic Tokenization (AAAI 2025 · similarity 0.87)
- Improving LLMs for Recommendation with Out-Of-Vocabulary Tokens (ICML 2025 · similarity 0.86)
- LLM-ESR: Large Language Models Enhancement for Long-tailed Sequential Recommendation (NeurIPS 2024 · similarity 0.85)
- CoRA: Collaborative Information Perception by Large Language Model’s Weights for Recommendation (AAAI 2025 · similarity 0.85)
- GeoMamba: Towards Multi-granular POI Recommendation with Geographical State Space Model (AAAI 2025 · similarity 0.85)
- GeoLLM: Extracting Geospatial Knowledge from Large Language Models (ICLR 2024 · similarity 0.85)
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.
