AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

HyCoRA: Hyper-Contrastive Role-Adaptive Learning for Role-Playing

Shihao Yang, Zhicong Lu, Yong Yang, Bo Lv, Yang Shen, Nayu Liu

AAAI 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
add hyper-contrastive learning (HCL) (ablation)HyCoRA on ChatGLM2-6B · RoleBench Chinese, general and specificBLEU-4: general 34.64 -> 35.09; specific 12.65 -> 13.40 when HCL addedImproved
use role-specific matrices for both A and B instead of shared B (ablation)Hyper-Half LoRA · RoleBench general, ChatGLM2-6BROUGE-L: 50.69 -> 48.89Hurt
share matrix A and make B role-specific (swap roles of A and B) (ablation)Hyper-Half LoRA · RoleBench general, ChatGLM2-6BROUGE-L: 50.69 -> 49.78Hurt
share matrix A and make B role-specific (ablation)Hyper-Half LoRA · RoleBench specific, ChatGLM2-6BBLEU-4: 13.40 -> 12.74Hurt
use HyCoRA instead of RoleLLM shared LoRAQwen2-7B, Chinese specific · RoleBench Chinese, specificROUGE-2: 20.63 -> 20.50 (Qwen2); 18.66 -> 19.78 (ChatGLM2)Mixed
use HyCoRA instead of shared LoRA in Chinese scenarioChatGLM2-6B, Qwen2-7B, Qwen2.5-7B · RoleBench ChineseBLEU / ROUGE: ChatGLM2 specific BLEU 12.09 -> 13.40Improved
use HyCoRA instead of shared LoRA in English scenarioLLaMA-2-7B · RoleBench EnglishBLEU: general 25.50 -> 26.26Improved
use HyCoRA instead of independent LoRA per roleQwen2-7B · RoleBench Chinese per roleROUGE-L: Role-1 general 50.30 -> 53.58Improved
use HyCoRA instead of independent LoRA on Role-5 specificQwen2-7B · RoleBench Chinese specific Role-5ROUGE-L: 31.97 -> 30.88Hurt
use HyCoRA instead of LoRA, GPT-4 judgedChatGLM2-6B, Qwen2-7B, Qwen2.5-7B, LLaMA-2-7B · RoleBench GPT-4 evaluationavg win rate: ChatGLM2 0.36 -> 0.56Improved
use HyCoRA instead of LoRA on LLaMA-2-7B Role KnowledgeLLaMA-2-7B · RoleBench GPT-4 RKwin rate: 0.38 -> 0.37No clear effect
use HyCoRA instead of independent LoRAtrainable parameters · 5 rolestrainable parametersImproved

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