Mercity/Memory_embedding

该数据集是一个包含411,137个高质量三元组的数据集，设计用于训练上下文感知的记忆检索和会话AI系统的对比学习模型。每个三元组由用户消息、相关记忆和上下文不相关的硬负记忆组成，源自现实场景。数据集具有丰富的上下文元数据，覆盖了20多个主题，包括个人项目、健康、人际关系、工作、爱好等，并采用了硬负样本挖掘技术。数据集适用于记忆增强的会话AI、对比学习、语义搜索、个性化AI助手和硬负样本挖掘研究等场景。

This dataset consists of 411,137 high-quality triplets designed for training contrastive learning models for context-aware memory retrieval and conversational AI systems. Each triplet includes a user message, relevant memories, and contextually irrelevant hard negative memories from realistic scenarios. The dataset features rich contextual metadata, diverse scenarios covering personal projects, health, relationships, work, hobbies, and more, with hard negative mining and multiple connection types. It is suitable for use cases such as memory-augmented conversational AI, contrastive learning, semantic search, personalized AI assistants, and hard negative mining research.