aaekay/pubmed-atlas

PubMed Atlas是一个生物医学文献数据集，基于28,460,827篇PubMed摘要的嵌入（来自aaekay/pubmed-embeddings发布），使用三种编码器（NeuML/pubmedbert-base-embeddings、Qwen/Qwen3-Embedding-0.6B和BAAI/bge-m3）独立计算。数据集提供层次聚类结果、MeSH派生标签、Qwen3.5-27B生成的主题命名，以及基于图谱的研究空白排名。它包含聚类分配、集群统计信息和研究空白指标，旨在支持生物医学文本的特征提取、分类和发现任务。数据集不重新分发PubMed文本，仅包含派生数据。

PubMed Atlas is a biomedical literature dataset based on embeddings of all 28,460,827 PubMed abstracts from the aaekay/pubmed-embeddings release, computed independently for three encoders: NeuML/pubmedbert-base-embeddings, Qwen/Qwen3-Embedding-0.6B, and BAAI/bge-m3. It provides hierarchical clustering, MeSH-derived labels, Qwen3.5-27B-generated topic names, and atlas-driven research-void rankings. The dataset includes cluster assignments, cluster statistics, and research void metrics, designed to support feature extraction, classification, and discovery tasks in biomedical text analysis. No PubMed text is redistributed; only derived data is included.