Research map / Large language models / Alignment and preferences
Causal Direct Preference Optimization for Distributionally Robust Generative Recommendation
Chu Zhao, Enneng Yang, Jianzhe Zhao, Guibing Guo
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 |
|---|---|---|---|
| use CausalDPO instead of strongest baseline SPRec | DPO-based LLM recommender · Yelp2018 (popularity shift) | HR@10 (%): 1.02 -> 1.53 | Improved |
| use CausalDPO instead of strongest baseline | DPO-based LLM recommender · ML-10M (temporal shift) | NDCG@20 (%): 0.38 -> 0.45 | Improved |
| use CausalDPO instead of baselines | DPO-based LLM recommender · Book-Crossing (exposure shift) | HR@10 (%): 0.59 -> 0.64 | Improved |
| add backdoor adjustment via soft clustering and MMD invariance regularization to DPO | DPO · three OOD datasets | HR/NDCG | Improved |
| skip SFT and run preference alignment directly (ablation) | CausalDPO · ablation datasets (Table 8) | HR/NDCG | Hurt |
| remove CausalDPO alignment, keep SFT only (ablation) | CausalDPO pipeline · ablation datasets (Table 8) | HR/NDCG | Improved |
| use CausalDPO instead of DPO | DPO · Book-Crossing | training time per epoch / accuracy: 2482 s -> 2971 s per epoch (+19.70%) | Mixed |
| use CausalDPO as backbone for SimPO/CPO variants (ablation) | SimPO, CPO · OOD recommendation (Table 5) | OOD performance | Improved |
| use CausalDPO instead of baselines under IID data | CausalDPO · IID (Table 3) | HR/NDCG | No clear effect |
Closest papers by meaning
- Mult-DPO: Multinomial Direct Preference Optimization for Recommender Systems (NeurIPS 2026 · similarity 0.89)
- Preference Diffusion for Recommendation (ICLR 2025 · similarity 0.89)
- On Softmax Direct Preference Optimization for Recommendation (NeurIPS 2024 · similarity 0.88)
- Less is More: Improving LLM Alignment via Preference Data Selection (NeurIPS 2025 · similarity 0.87)
- Provably Robust DPO: Aligning Language Models with Noisy Feedback (ICML 2024 · similarity 0.87)
- InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment (CVPR 2025 · similarity 0.87)
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.
