Research map / Large language models / Retrieval and knowledge
Diagnostic-Guided Dynamic Profile Optimization for LLM-based User Simulators in Sequential Recommendation
Hongyang Liu, Zhu Sun, Tianjun Wei, Yan Wang, Jiajie Zhu, Xinghua Qu
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 |
|---|---|---|---|
| iterative step-by-step profile optimization instead of one-shot (ablation) | DGDPO · Books | F1: 0.2753 -> 0.3235 | Improved |
| use specialized fine-tuned diagnostic LLM instead of a general-purpose LLM (ablation) | DGDPO · Books | F1: 0.2760 -> 0.3235 | Improved |
| add defect-specific fine-tuning of the diagnostic LLM (ablation) | DGDPO · Books | F1: 0.2885 -> 0.3235 | Improved |
| add domain-adaptive pre-training of the diagnostic LLM (ablation) | DGDPO · Books | F1: 0.3183 -> 0.3235 | Improved |
| use specialized Qwen3-0.6B diagnostic module vs prompting GPT-4o-mini for defect identification | diagnostic module · profile defect identification (average over datasets) | accuracy (%): 62.78% -> 92.20% | Improved |
| scale specialized diagnostic LLM from Qwen3-0.6B to Qwen3-1.7B (ablation) | DGDPO · Books | F1: 0.3235 -> 0.3170 | Hurt |
| raise initial/optimization split ratio alpha from 0.6 to 0.8 (ablation) | DGDPO · Books | F1: 0.3235 -> 0.3001 | Hurt |
| lower split ratio alpha from 0.6 to 0.5 (ablation) | DGDPO · Books | F1: 0.3235 -> 0.3212 | Hurt |
| update profile during multi-round interaction with ground truth item (ablation) | DGDPO with sequential recommenders · Books (multi-round, Precision) | Precision | Improved |
| update profile during multi-round interaction without ground truth (simulator's own choice) with FPMC (ablation) | DGDPO with FPMC · Books (multi-round, Precision) | Precision | Hurt |
| use DGDPO over strongest baseline (AgentCF) | user simulator · Books | Precision: 0.1598 -> 0.2058 | Improved |
Closest papers by meaning
- LLM-Powered User Simulator for Recommender System (AAAI 2025 · similarity 0.89)
- LLM-ESR: Large Language Models Enhancement for Long-tailed Sequential Recommendation (NeurIPS 2024 · similarity 0.87)
- LLM4RSR: Large Language Models as Data Correctors for Robust Sequential Recommendation (AAAI 2025 · similarity 0.87)
- SLMRec: Distilling Large Language Models into Small for Sequential Recommendation (ICLR 2025 · similarity 0.87)
- Can LLMs Outshine Conventional Recommenders? A Comparative Evaluation (NeurIPS 2025 · similarity 0.87)
- LLMRG: Improving Recommendations through Large Language Model Reasoning Graphs (AAAI 2024 · similarity 0.86)
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.
