Research map / Vision-language and multimodal / Multimodal LLMs
DiMA: Distinguishing Resident and Tourist Preferences via Multi-Modal LLM Alignment for Out-of-Town Cross-Domain Recommendation
Fan Zhang, Jinpeng Chen, Tao Wang, Huan Li, Senzhang Wang, Feifei Kou, Ye Ji, Kaimin Wei, Zhenye Yang
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 DiMA, a multi-modal LLM alignment approach for out-of-town cross-domain recommendation that distinguishes resident and tourist preferences.
Problem
Out-of-town recommendation for users in unfamiliar cities faces cross-modal preference reasoning (images vs. text are hard to compare) and preference deviation between a user's resident and tourist preferences.
Approach
Aligns multi-modal (image and text) preference signals with an LLM and explicitly separates resident from tourist preference so simple transfer does not suffer from deviation; only the abstract is available, so finer mechanism details are not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| distill teacher CoT reasoning text (not only final ranking) (ablation) | DiMA · Foursquare KDDC (avg of 6 tasks) | NDCG@5: 0.384 -> 0.435 | Improved |
| add DPO stage after SFT (ablation) | DiMA · Foursquare KDDC (avg of 6 tasks) | NDCG@5: 0.412 -> 0.435 | Improved |
| add collaborative filtering scores to prompt (ablation) | DiMA · Foursquare KDDC (avg of 6 tasks) | NDCG@5: 0.422 -> 0.435 | Improved |
| add multi-modal tags to prompt (ablation) | DiMA · Foursquare KDDC (avg of 6 tasks) | NDCG@5: 0.432 -> 0.435 | Improved |
| add auxiliary travel history (3-city vs 2-city prompt) (ablation) | DiMA student · Foursquare KDDC LA→NY→SF | NDCG@5: 0.442 -> 0.457 | Improved |
| add auxiliary travel history (3-city vs 2-city prompt) (ablation) | DiMA student · Foursquare CAPTOR LA→NY→SF | NDCG@5: 0.396 -> 0.414 | Improved |
| SFT+DPO student replaces teacher DeepSeek-R1 (3-city) | DiMA · Foursquare KDDC SF→LA→NY | NDCG@5: 0.353 -> 0.372 | Improved |
| student DiMA vs Teacher-3C zero-shot on Yelp | DiMA · Yelp LA→NY→SF | NDCG@5: 0.040 -> 0.035 | Hurt |
| use larger student model (Qwen3-4B vs smaller) (ablation) | DiMA · Foursquare KDDC | NDCG/MAP | Improved |
Closest papers by meaning
- SPOT-Trip: Dual-Preference Driven Out-of-Town Trip Recommendation (NeurIPS 2025 · similarity 0.88)
- Enhancing Long-and Short-Term Representations for Next POI Recommendations via Frequency and Hierarchical Contrastive Learning (AAAI 2025 · similarity 0.85)
- Harnessing Multimodal Large Language Models for Multimodal Sequential Recommendation (AAAI 2025 · similarity 0.84)
- TRACE: Tourism Recommendation with Accountable Citation Evidence (NeurIPS 2026 · similarity 0.84)
- Tree of Preferences for Diversified Recommendation (NeurIPS 2025 · similarity 0.84)
- Preference Diffusion for Recommendation (ICLR 2025 · similarity 0.84)
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.
