NEAR：面向记忆检索的邻域感知残差方法
原标题：NEAR: Neighborhood-Aware Residuals for Memory Retrieval
AI 摘要
论文提出 NEAR（Neighborhood-Aware Residuals），一种无需训练的框架，用于记忆检索。该方法利用相似记忆记录之间的共享内容构建局部参考方向，为查询和记忆生成投影残差表示，与原始嵌入结合后进行单次近似最近邻搜索。在八个长上下文检索基准和九个嵌入模型上，NEAR 一致提升了 Recall@10 和 NDCG@10，并在 LoCoMo 上使用 Qwen3-Embedding-4B 时将问答准确率提升 9.6 个百分点。
正文节选
NEAR: Neighborhood-Aware Residuals for Memory Retrieval Abstract Memory retrieval aims to identify evidence from stored records to answer a query. Stored records with different factual details may share topics, entities, or events. Such shared contents would make records without the required evidence more likely to receive high similarity scores and outrank the correct evidence. In this paper, we use the information shared by these records to construct local references for query–memory matching.