AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

RecCocktail: A Generalizable and Efficient Framework for LLM-Based Recommendation

Min Hou, Chenxi Bai, Le Wu, Hao Liu, Kai Zhang, Weiwen Liu, Richang Hong, Ruiming Tang, Meng Wang

AAAI 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 4 citations (OpenAlex)

Read the paper

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 changedWhereMeasuredEffect
merge domain-general and domain-specific LoRA (entropy-guided) instead of domain-specific only (ablation)RecCocktail (Qwen2-7B) · Beauty warm-startNDCG@1: 0.4079 -> 0.4132Improved
merge in domain-specific LoRA with base spirit instead of general only (ablation)RecCocktail (Qwen2-7B) · Beauty warm-startNDCG@1: 0.3081 -> 0.4132Improved
merge in base spirit LoRA instead of specific only (ablation)RecCocktail (Qwen2-7B) · Beauty cold-startNDCG@1: 0.1603 -> 0.1825Improved
entropy-guided adaptive merging instead of simple weight average (ablation)RecCocktail (Qwen2-7B) · MovieLens-1M cold-startNDCG@1: 0.1636 -> 0.1818Improved
entropy-guided adaptive merging instead of simple weight average (ablation)RecCocktail (Qwen2-7B) · Beauty warm-startNDCG@1: 0.3722 -> 0.4132Improved
weight average merging instead of domain-specific LoRA only (ablation)RecCocktail-WA · Sports warm-startNDCG@1: 0.3735 -> 0.3410Hurt
merge LoRAs instead of second-round fine-tuning of base LoRA (ablation)RecCocktail (Qwen2-7B) · MovieLens-1M few-shot 10% warmNDCG@1: 0.0704 -> 0.5353Improved
use RecCocktail instead of TALLRec in few-shotRecCocktail (Qwen2-7B) · MovieLens-1M few-shot 10% warmNDCG@1: 0.3957 -> 0.5353Improved
swap backbone to Llama-3.1-8B and merge (ablation)RecCocktail (Llama-3.1-8B) · Sports cold-startNDCG@1: 0.1538 -> 0.1587Improved
use more unlabeled test samples (50 vs 100 vs 200) for entropy merging (ablation)RecCocktail · Beauty/Toys/SportsNDCG@1No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy