PET-BENCH
收藏资源简介:
PET-BENCH是一个专门为评估大型语言模型(LLM)在社交网络服务(SNS)中的虚拟宠物陪伴能力而设计的基准数据集。它包含超过7,500个交互实例,旨在模拟复杂的宠物行为,并涵盖智能日程安排、基于记忆的对话和心理对话等多种任务。数据集通过模拟真实宠物与主人的互动,包括情感表达、日常活动、记忆摘要和意图注释等,以评估LLM在虚拟宠物陪伴场景下的表现。PET-BENCH旨在推动LLM在虚拟宠物陪伴领域的应用和发展,为研究情感沉浸式人宠互动提供基础资源。
PET-BENCH is a benchmark dataset specifically developed to evaluate the virtual pet companionship capabilities of Large Language Models (LLMs) in Social Network Services (SNS). It contains over 7,500 interaction instances, which are designed to simulate complex pet behaviors and cover a variety of tasks including intelligent scheduling, memory-based dialogue, and psychological dialogue. The dataset simulates real pet-owner interactions, incorporating emotional expressions, daily activities, memory summaries, and intent annotations, to assess the performance of LLMs in virtual pet companionship scenarios. PET-BENCH aims to advance the application and development of LLMs in the field of virtual pet companionship, serving as a foundational resource for research on emotionally immersive human-pet interaction.




